Train Sleep Detection Model¶

Date created: 2024/10/02

Last Modified: 2024/10/02

Description: Train a simple wrist-based sleep detection model using accelerometer data.

Overview¶

In this guide, we will train a small TCN network to detect sleep and wake stages using accelerometer data collected from the wrist.

Input

• Sensor: IMU
• Location: Wrist
• Sampling Rate: 0.2 Hz
• Frame Size: 60 seconds

Class Mapping

Identify activity into one of two categories: SLEEP, AWAKE.

Base Class	Target Class	Label
0-WAKE	0	WAKE
1-SLEEP	1	SLEEP

Datasets

• CMIDSS: The Child Mind Institute - Detect Sleep States (CMIDSS) dataset comprises 300 subjects with over 500 multi-day recordings of wrist-worn accelerometer data annotated with two event types: onset, the beginning of sleep, and wakeup, the end of sleep.

Setup¶

In [1]:

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
import IPython
import contextlib
import tempfile
from pathlib import Path
import keras
import neuralspot_edge as nse
import sleepkit as sk
import matplotlib.pyplot as plt
import plotly.io as pio

import os os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' import IPython import contextlib import tempfile from pathlib import Path import keras import neuralspot_edge as nse import sleepkit as sk import matplotlib.pyplot as plt import plotly.io as pio

/workspaces/sleepkit/.venv/lib/python3.11/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm

In [2]:

# Be sure to set the dataset path to the correct location
datasets_dir = Path(os.getenv('SK_DATASET_PATH', '../../datasets'))

plot_theme = sk.utils.dark_theme
nse.utils.silence_tensorflow()
sk.utils.setup_plotting(plot_theme)
logger = nse.utils.setup_logger(__name__)

# Be sure to set the dataset path to the correct location datasets_dir = Path(os.getenv('SK_DATASET_PATH', '../../datasets')) plot_theme = sk.utils.dark_theme nse.utils.silence_tensorflow() sk.utils.setup_plotting(plot_theme) logger = nse.utils.setup_logger(__name__)

Configure datasets¶

We are going to train our model using the [CMIDSS Dataset(https://ambiqai.github.io/sleepkit/datasets/cmidss/). This dataset uses the slug cmidss within SleepKit. We will download the dataset if it is not already available.

In [3]:

datasets = [sk.NamedParams(
    name="cmidss",
    params=dict(
        path=datasets_dir / "cmidss",
    )
)]

datasets = [sk.NamedParams( name="cmidss", params=dict( path=datasets_dir / "cmidss", ) )]

Target classes¶

For this task, we are going to simply classify the data into two classes: SLEEP and AWAKE.

In [4]:

class_map = {
    sk.SleepStage.wake: 0,
    sk.SleepStage.stage1: 1,
    sk.SleepStage.stage2: 1,
    sk.SleepStage.stage3: 1,
    sk.SleepStage.stage4: 1,
    sk.SleepStage.rem: 1,
}

class_names = ["WAKE", "SLEEP"]

class_map = { sk.SleepStage.wake: 0, sk.SleepStage.stage1: 1, sk.SleepStage.stage2: 1, sk.SleepStage.stage3: 1, sk.SleepStage.stage4: 1, sk.SleepStage.rem: 1, } class_names = ["WAKE", "SLEEP"]

Feature set¶

From the dataset, we will create a feature set using the FS-W-A-5 features. This feature set computes 5 features over 60-second windows captured from the accelerometer sensor collected on the wrist. The CMIDSS dataset already provides accelerometer averaged over 5 secods (i.e. Fs=0.2 Hz). Therefore, we will use a frame size of 12 to capture 60 seconds of data (i.e. 6 samples at 0.2 Hz) with a 50% overlap.

In [5]:

feature = dict(
    name="FS-W-A-5",
    sampling_rate=0.2,
    frame_size=12,
    loader="hdf5",
    feat_key="features",
    label_key="detect_labels",
    mask_key="mask",
    feat_cols=None,
    save_path=datasets_dir / "store" / "fs-w-a-5-60",
    params={},
)

feature = dict( name="FS-W-A-5", sampling_rate=0.2, frame_size=12, loader="hdf5", feat_key="features", label_key="detect_labels", mask_key="mask", feat_cols=None, save_path=datasets_dir / "store" / "fs-w-a-5-60", params={}, )

Define TCN model architecture¶

For this task, we are going to leverage a customized TCN model architecture that is smaller and can handle 1D signals. The model consists of 4 TCN blocks with a depth of 1. Each block leverages dilated depthwise-separable convolutions along with inverted expansion and squeeze and excitation layers. The model is followed by a 1D convolutional layer and a final dense layer for regression. Unlike vision tasks, we leverage larger kernel sizes and strides to capture temporal dependencies in the signal.

In [6]:

architecture = sk.NamedParams(
    name="tcn",
    params=dict(
        input_kernel=[1, 5],
        input_norm="batch",
        blocks=[
            dict(depth=1, branch=1, filters=16, kernel=(1, 5), dilation=[1, 1], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch"),
            dict(depth=1, branch=1, filters=32, kernel=(1, 5), dilation=[1, 2], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch"),
            dict(depth=1, branch=1, filters=48, kernel=(1, 5), dilation=[1, 4], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch"),
            dict(depth=1, branch=1, filters=64, kernel=(1, 5), dilation=[1, 8], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch")
        ],
        output_kernel=(1, 5),
        include_top=True,
        use_logits=True,
        model_name="tcn"
    )
)

architecture = sk.NamedParams( name="tcn", params=dict( input_kernel=[1, 5], input_norm="batch", blocks=[ dict(depth=1, branch=1, filters=16, kernel=(1, 5), dilation=[1, 1], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch"), dict(depth=1, branch=1, filters=32, kernel=(1, 5), dilation=[1, 2], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch"), dict(depth=1, branch=1, filters=48, kernel=(1, 5), dilation=[1, 4], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch"), dict(depth=1, branch=1, filters=64, kernel=(1, 5), dilation=[1, 8], dropout=0.10, ex_ratio=1, se_ratio=4, norm="batch") ], output_kernel=(1, 5), include_top=True, use_logits=True, model_name="tcn" ) )

Task configuration¶

Here we provide the complete configuration for the task. This includes the dataset configuration, features, model architecture, and training parameters.

In [7]:

params = sk.TaskParams(
    name="sk-detect",
    job_dir=Path(tempfile.gettempdir()) / "sk-detect",
    verbose=1,

    datasets=datasets,

    feature=feature,
    sampling_rate=0.0083333,
    frame_size=240,

    num_classes=len(class_names),
    class_map=class_map,
    class_names=class_names,

    samples_per_subject=100,
    val_samples_per_subject=100,
    test_samples_per_subject=50,

    val_size=4000,
    test_size=2500,

    val_subjects=0.20,
    batch_size=128,
    buffer_size=10000,
    epochs=200,
    steps_per_epoch=25,
    val_steps_per_epoch=25,
    val_metric="loss",
    lr_rate=1e-3,
    lr_cycles=1,
    label_smoothing=0,

    test_metric="f1",
    test_metric_threshold=0.02,
    tflm_var_name="sk_detect_flatbuffer",
    tflm_file="sk_detect_flatbuffer.h",

    backend="pc",
    display_report=False,

    model_file="model.keras",
    use_logits=False,
    architecture=architecture
)

params = sk.TaskParams( name="sk-detect", job_dir=Path(tempfile.gettempdir()) / "sk-detect", verbose=1, datasets=datasets, feature=feature, sampling_rate=0.0083333, frame_size=240, num_classes=len(class_names), class_map=class_map, class_names=class_names, samples_per_subject=100, val_samples_per_subject=100, test_samples_per_subject=50, val_size=4000, test_size=2500, val_subjects=0.20, batch_size=128, buffer_size=10000, epochs=200, steps_per_epoch=25, val_steps_per_epoch=25, val_metric="loss", lr_rate=1e-3, lr_cycles=1, label_smoothing=0, test_metric="f1", test_metric_threshold=0.02, tflm_var_name="sk_detect_flatbuffer", tflm_file="sk_detect_flatbuffer.h", backend="pc", display_report=False, model_file="model.keras", use_logits=False, architecture=architecture )

Load detect task¶

SleepKit provides a TaskFactory that includes a number ready-to-use tasks. Each task provides methods for training, evaluating, exporting, and demoing. We will grab the detect task and configure it for our use case.

In [8]:

task = sk.TaskFactory.get("detect")

task = sk.TaskFactory.get("detect")

Download the datasets¶

We will download the datasets using the sleepkit library. If already downloaded, this step will be skipped.

In [17]:

task.download(params=params)

task.download(params=params)

Generate the features¶

Next, we will generate the features from the given dataset. The features will be generated using the fs_w_a_5 feature set.

Once the command finishes, the feature set will be saved in the feature.save_path directory. These features will be stored in HDF5 files with one file per subject. Each HDF5 file will include the following entries:

• /features: Time x Feature tensor (fp32). Features are computed over windows of sensor data.
• /mask: Time x Mask tensor (bool). Mask indicates valid feature values.
• /detect_labels: Time x Label (int). Labels are awake/sleep.

In [18]:

task.feature(params=params)

task.feature(params=params)

Gen features for cmidss: 100%|██████████| 277/277 [01:59<00:00,  2.32it/s]

Visualize the model¶

Lets quickly instantiate and visualize the model.

In [19]:

model = nse.models.tcn.TcnModel.model_from_params(
    inputs=keras.Input(shape=(params.frame_size, 5), name="inputs"),
    params=architecture.params,
    num_classes=len(class_names)
)
model.summary(layer_range=('inputs', model.layers[10].name))

model = nse.models.tcn.TcnModel.model_from_params( inputs=keras.Input(shape=(params.frame_size, 5), name="inputs"), params=architecture.params, num_classes=len(class_names) ) model.summary(layer_range=('inputs', model.layers[10].name))

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1727880897.880315    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880897.954085    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880897.954212    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880897.956184    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880897.956276    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880897.956323    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880898.013209    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880898.013298    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727880898.013358    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355

Model: "functional"

┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓
┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃
┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩
│ inputs (InputLayer) │ (None, 240, 5)    │          0 │ -                 │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ reshape (Reshape)   │ (None, 1, 240, 5) │          0 │ inputs[0][0]      │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ ENC.CN              │ (None, 1, 240, 5) │         25 │ reshape[0][0]     │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ ENC.BN              │ (None, 1, 240, 5) │         20 │ ENC.CN[0][0]      │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.DW.B1.CN      │ (None, 1, 240, 5) │         25 │ ENC.BN[0][0]      │
│ (DepthwiseConv2D)   │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.DW.B1.BN      │ (None, 1, 240, 5) │         20 │ B1.D1.DW.B1.CN[0… │
│ (BatchNormalizatio… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.DW.ACT        │ (None, 1, 240, 5) │          0 │ B1.D1.DW.B1.BN[0… │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.SE.pool       │ (None, 1, 1, 5)   │          0 │ B1.D1.DW.ACT[0][… │
│ (GlobalAveragePool… │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.SE.sq         │ (None, 1, 1, 1)   │          6 │ B1.D1.SE.pool[0]… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.SE.sq.act     │ (None, 1, 1, 1)   │          0 │ B1.D1.SE.sq[0][0] │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.SE.ex         │ (None, 1, 1, 5)   │         10 │ B1.D1.SE.sq.act[… │
│ (Conv2D)            │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.SE.ex.act     │ (None, 1, 1, 5)   │          0 │ B1.D1.SE.ex[0][0] │
│ (Activation)        │                   │            │                   │
├─────────────────────┼───────────────────┼────────────┼───────────────────┤
│ B1.D1.SE.ex.mul     │ (None, 1, 240, 5) │          0 │ B1.D1.DW.ACT[0][… │
│ (Multiply)          │                   │            │ B1.D1.SE.ex.act[… │
└─────────────────────┴───────────────────┴────────────┴───────────────────┘

 Total params: 9,364 (36.58 KB)

 Trainable params: 8,832 (34.50 KB)

 Non-trainable params: 532 (2.08 KB)

Train the model¶

At this point, we can train the model from the generated feature set for the sleep detect task. The model will be trained for 200 epochs with a batch size of 128 and a learning rate of 1e-3. The model will be fed a frame_size of 240 samples which equates to 120 minutes.

Using the task configuration, we will train the model on the dataset.

In [20]:

task.train(params)

task.train(params)

Epoch 1/200

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1727880905.268682    6547 service.cc:146] XLA service 0x7cb4a4011fd0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1727880905.268702    6547 service.cc:154]   StreamExecutor device (0): NVIDIA GeForce RTX 4090, Compute Capability 8.9

 7/25 ━━━━━━━━━━━━━━━━━━━━ 0s 26ms/step - acc: 0.5802 - f1: 0.6041 - iou: 0.3636 - loss: 0.4143

I0000 00:00:1727880911.515249    6547 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.

25/25 ━━━━━━━━━━━━━━━━━━━━ 13s 80ms/step - acc: 0.6424 - f1: 0.6496 - iou: 0.4014 - loss: 0.3955 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.3363
Epoch 2/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 35ms/step - acc: 0.7426 - f1: 0.7359 - iou: 0.4776 - loss: 0.3269 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.2995
Epoch 3/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.7694 - f1: 0.7657 - iou: 0.5360 - loss: 0.2839 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.2656
Epoch 4/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.7952 - f1: 0.7877 - iou: 0.5491 - loss: 0.2479 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.2366
Epoch 5/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.7991 - f1: 0.7946 - iou: 0.5685 - loss: 0.2184 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.2112
Epoch 6/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.8120 - f1: 0.8088 - iou: 0.6102 - loss: 0.1927 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1893
Epoch 7/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.8154 - f1: 0.8166 - iou: 0.6085 - loss: 0.1709 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1704
Epoch 8/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.8334 - f1: 0.8362 - iou: 0.6520 - loss: 0.1513 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1542
Epoch 9/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.8465 - f1: 0.8449 - iou: 0.6582 - loss: 0.1339 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1395
Epoch 10/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.8530 - f1: 0.8566 - iou: 0.6995 - loss: 0.1197 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1279
Epoch 11/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.8705 - f1: 0.8701 - iou: 0.7140 - loss: 0.1071 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1168
Epoch 12/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.8801 - f1: 0.8817 - iou: 0.7295 - loss: 0.0959 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.1073
Epoch 13/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.8821 - f1: 0.8822 - iou: 0.7343 - loss: 0.0866 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0996
Epoch 14/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9043 - f1: 0.9051 - iou: 0.7723 - loss: 0.0769 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0936
Epoch 15/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9053 - f1: 0.9056 - iou: 0.7832 - loss: 0.0701 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0872
Epoch 16/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9114 - f1: 0.9132 - iou: 0.7963 - loss: 0.0631 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0831
Epoch 17/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9196 - f1: 0.9200 - iou: 0.8157 - loss: 0.0569 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0792
Epoch 18/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9307 - f1: 0.9313 - iou: 0.8343 - loss: 0.0510 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0771
Epoch 19/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9224 - f1: 0.9230 - iou: 0.8160 - loss: 0.0484 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0732
Epoch 20/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9274 - f1: 0.9278 - iou: 0.8212 - loss: 0.0440 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0704
Epoch 21/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9152 - f1: 0.9169 - iou: 0.8028 - loss: 0.0418 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0626
Epoch 22/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9177 - f1: 0.9191 - iou: 0.8108 - loss: 0.0394 - val_acc: 0.7596 - val_f1: 0.6558 - val_iou: 0.3798 - val_loss: 0.0639
Epoch 23/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9278 - f1: 0.9286 - iou: 0.8268 - loss: 0.0357 - val_acc: 0.7594 - val_f1: 0.6563 - val_iou: 0.3803 - val_loss: 0.0605
Epoch 24/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9209 - f1: 0.9226 - iou: 0.8139 - loss: 0.0338 - val_acc: 0.7600 - val_f1: 0.6597 - val_iou: 0.3839 - val_loss: 0.0577
Epoch 25/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9236 - f1: 0.9249 - iou: 0.8217 - loss: 0.0318 - val_acc: 0.7676 - val_f1: 0.6847 - val_iou: 0.4114 - val_loss: 0.0487
Epoch 26/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - acc: 0.9201 - f1: 0.9216 - iou: 0.8130 - loss: 0.0305 - val_acc: 0.7635 - val_f1: 0.6727 - val_iou: 0.3979 - val_loss: 0.0540
Epoch 27/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9367 - f1: 0.9374 - iou: 0.8488 - loss: 0.0274 - val_acc: 0.7632 - val_f1: 0.6713 - val_iou: 0.3964 - val_loss: 0.0556
Epoch 28/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9260 - f1: 0.9274 - iou: 0.8260 - loss: 0.0270 - val_acc: 0.7709 - val_f1: 0.6933 - val_iou: 0.4214 - val_loss: 0.0478
Epoch 29/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9237 - f1: 0.9252 - iou: 0.8189 - loss: 0.0259 - val_acc: 0.7616 - val_f1: 0.6638 - val_iou: 0.3882 - val_loss: 0.0633
Epoch 30/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9338 - f1: 0.9349 - iou: 0.8392 - loss: 0.0240 - val_acc: 0.7889 - val_f1: 0.7352 - val_iou: 0.4745 - val_loss: 0.0435
Epoch 31/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9353 - f1: 0.9362 - iou: 0.8455 - loss: 0.0230 - val_acc: 0.7769 - val_f1: 0.7076 - val_iou: 0.4387 - val_loss: 0.0487
Epoch 32/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9331 - f1: 0.9341 - iou: 0.8339 - loss: 0.0219 - val_acc: 0.7825 - val_f1: 0.7210 - val_iou: 0.4557 - val_loss: 0.0451
Epoch 33/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9316 - f1: 0.9327 - iou: 0.8439 - loss: 0.0213 - val_acc: 0.7917 - val_f1: 0.7400 - val_iou: 0.4811 - val_loss: 0.0414
Epoch 34/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9280 - f1: 0.9290 - iou: 0.8377 - loss: 0.0215 - val_acc: 0.8168 - val_f1: 0.7842 - val_iou: 0.5459 - val_loss: 0.0352
Epoch 35/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9336 - f1: 0.9349 - iou: 0.8405 - loss: 0.0197 - val_acc: 0.8428 - val_f1: 0.8241 - val_iou: 0.6127 - val_loss: 0.0297
Epoch 36/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9314 - f1: 0.9327 - iou: 0.8417 - loss: 0.0194 - val_acc: 0.8045 - val_f1: 0.7640 - val_iou: 0.5153 - val_loss: 0.0356
Epoch 37/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9366 - f1: 0.9375 - iou: 0.8476 - loss: 0.0183 - val_acc: 0.7892 - val_f1: 0.7341 - val_iou: 0.4729 - val_loss: 0.0422
Epoch 38/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9399 - f1: 0.9412 - iou: 0.8570 - loss: 0.0173 - val_acc: 0.7822 - val_f1: 0.7176 - val_iou: 0.4511 - val_loss: 0.0469
Epoch 39/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9421 - f1: 0.9429 - iou: 0.8587 - loss: 0.0166 - val_acc: 0.8549 - val_f1: 0.8420 - val_iou: 0.6459 - val_loss: 0.0287
Epoch 40/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9376 - f1: 0.9383 - iou: 0.8513 - loss: 0.0166 - val_acc: 0.8677 - val_f1: 0.8584 - val_iou: 0.6769 - val_loss: 0.0247
Epoch 41/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9298 - f1: 0.9310 - iou: 0.8284 - loss: 0.0170 - val_acc: 0.8873 - val_f1: 0.8860 - val_iou: 0.7365 - val_loss: 0.0226
Epoch 42/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9367 - f1: 0.9378 - iou: 0.8478 - loss: 0.0164 - val_acc: 0.8938 - val_f1: 0.8963 - val_iou: 0.7630 - val_loss: 0.0201
Epoch 43/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9432 - f1: 0.9441 - iou: 0.8625 - loss: 0.0151 - val_acc: 0.8997 - val_f1: 0.8997 - val_iou: 0.7655 - val_loss: 0.0194
Epoch 44/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9439 - f1: 0.9446 - iou: 0.8636 - loss: 0.0147 - val_acc: 0.8770 - val_f1: 0.8682 - val_iou: 0.6950 - val_loss: 0.0243
Epoch 45/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9415 - f1: 0.9421 - iou: 0.8580 - loss: 0.0149 - val_acc: 0.9113 - val_f1: 0.9121 - val_iou: 0.7922 - val_loss: 0.0189
Epoch 46/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9383 - f1: 0.9394 - iou: 0.8530 - loss: 0.0150 - val_acc: 0.9101 - val_f1: 0.9107 - val_iou: 0.7888 - val_loss: 0.0178
Epoch 47/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9337 - f1: 0.9348 - iou: 0.8452 - loss: 0.0155 - val_acc: 0.9075 - val_f1: 0.9078 - val_iou: 0.7823 - val_loss: 0.0178
Epoch 48/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9431 - f1: 0.9439 - iou: 0.8686 - loss: 0.0144 - val_acc: 0.9060 - val_f1: 0.9042 - val_iou: 0.7716 - val_loss: 0.0182
Epoch 49/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9450 - f1: 0.9457 - iou: 0.8727 - loss: 0.0136 - val_acc: 0.9094 - val_f1: 0.9120 - val_iou: 0.7958 - val_loss: 0.0171
Epoch 50/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9408 - f1: 0.9418 - iou: 0.8567 - loss: 0.0139 - val_acc: 0.9075 - val_f1: 0.9099 - val_iou: 0.7910 - val_loss: 0.0169
Epoch 51/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9387 - f1: 0.9399 - iou: 0.8487 - loss: 0.0137 - val_acc: 0.9144 - val_f1: 0.9158 - val_iou: 0.8010 - val_loss: 0.0165
Epoch 52/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 17ms/step - acc: 0.9346 - f1: 0.9350 - iou: 0.8463 - loss: 0.0142 - val_acc: 0.9045 - val_f1: 0.9081 - val_iou: 0.7897 - val_loss: 0.0169
Epoch 53/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9390 - f1: 0.9395 - iou: 0.8490 - loss: 0.0145 - val_acc: 0.9114 - val_f1: 0.9135 - val_iou: 0.7979 - val_loss: 0.0161
Epoch 54/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9416 - f1: 0.9423 - iou: 0.8567 - loss: 0.0129 - val_acc: 0.8999 - val_f1: 0.9037 - val_iou: 0.7811 - val_loss: 0.0175
Epoch 55/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9421 - f1: 0.9428 - iou: 0.8567 - loss: 0.0128 - val_acc: 0.9144 - val_f1: 0.9147 - val_iou: 0.7969 - val_loss: 0.0172
Epoch 56/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9422 - f1: 0.9431 - iou: 0.8513 - loss: 0.0128 - val_acc: 0.8573 - val_f1: 0.8423 - val_iou: 0.6448 - val_loss: 0.0300
Epoch 57/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9484 - f1: 0.9492 - iou: 0.8688 - loss: 0.0116 - val_acc: 0.9030 - val_f1: 0.9036 - val_iou: 0.7743 - val_loss: 0.0177
Epoch 58/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9424 - f1: 0.9430 - iou: 0.8671 - loss: 0.0133 - val_acc: 0.8995 - val_f1: 0.8980 - val_iou: 0.7596 - val_loss: 0.0184
Epoch 59/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9425 - f1: 0.9432 - iou: 0.8596 - loss: 0.0123 - val_acc: 0.9223 - val_f1: 0.9233 - val_iou: 0.8162 - val_loss: 0.0146
Epoch 60/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9324 - f1: 0.9329 - iou: 0.8439 - loss: 0.0148 - val_acc: 0.9216 - val_f1: 0.9219 - val_iou: 0.8120 - val_loss: 0.0146
Epoch 61/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9506 - f1: 0.9511 - iou: 0.8775 - loss: 0.0112 - val_acc: 0.9112 - val_f1: 0.9147 - val_iou: 0.8037 - val_loss: 0.0166
Epoch 62/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9426 - f1: 0.9431 - iou: 0.8659 - loss: 0.0126 - val_acc: 0.9232 - val_f1: 0.9242 - val_iou: 0.8184 - val_loss: 0.0143
Epoch 63/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9486 - f1: 0.9494 - iou: 0.8724 - loss: 0.0116 - val_acc: 0.9201 - val_f1: 0.9226 - val_iou: 0.8179 - val_loss: 0.0151
Epoch 64/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9324 - f1: 0.9332 - iou: 0.8377 - loss: 0.0138 - val_acc: 0.8908 - val_f1: 0.8965 - val_iou: 0.7708 - val_loss: 0.0205
Epoch 65/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9397 - f1: 0.9407 - iou: 0.8572 - loss: 0.0123 - val_acc: 0.9289 - val_f1: 0.9304 - val_iou: 0.8330 - val_loss: 0.0146
Epoch 66/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9468 - f1: 0.9475 - iou: 0.8720 - loss: 0.0112 - val_acc: 0.9133 - val_f1: 0.9168 - val_iou: 0.8077 - val_loss: 0.0149
Epoch 67/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9407 - f1: 0.9418 - iou: 0.8597 - loss: 0.0122 - val_acc: 0.9174 - val_f1: 0.9200 - val_iou: 0.8124 - val_loss: 0.0146
Epoch 68/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9461 - f1: 0.9464 - iou: 0.8674 - loss: 0.0110 - val_acc: 0.9084 - val_f1: 0.9123 - val_iou: 0.7996 - val_loss: 0.0154
Epoch 69/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9291 - f1: 0.9302 - iou: 0.8383 - loss: 0.0131 - val_acc: 0.9256 - val_f1: 0.9274 - val_iou: 0.8270 - val_loss: 0.0139
Epoch 70/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9466 - f1: 0.9473 - iou: 0.8730 - loss: 0.0117 - val_acc: 0.9168 - val_f1: 0.9195 - val_iou: 0.8119 - val_loss: 0.0143
Epoch 71/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9443 - f1: 0.9452 - iou: 0.8654 - loss: 0.0116 - val_acc: 0.9243 - val_f1: 0.9264 - val_iou: 0.8254 - val_loss: 0.0137
Epoch 72/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9486 - f1: 0.9494 - iou: 0.8737 - loss: 0.0117 - val_acc: 0.9260 - val_f1: 0.9276 - val_iou: 0.8269 - val_loss: 0.0143
Epoch 73/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9464 - f1: 0.9471 - iou: 0.8724 - loss: 0.0114 - val_acc: 0.9246 - val_f1: 0.9244 - val_iou: 0.8163 - val_loss: 0.0142
Epoch 74/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9444 - f1: 0.9449 - iou: 0.8658 - loss: 0.0116 - val_acc: 0.9275 - val_f1: 0.9289 - val_iou: 0.8292 - val_loss: 0.0134
Epoch 75/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9477 - f1: 0.9484 - iou: 0.8706 - loss: 0.0107 - val_acc: 0.9225 - val_f1: 0.9249 - val_iou: 0.8228 - val_loss: 0.0137
Epoch 76/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9462 - f1: 0.9469 - iou: 0.8717 - loss: 0.0121 - val_acc: 0.9197 - val_f1: 0.9224 - val_iou: 0.8178 - val_loss: 0.0141
Epoch 77/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9492 - f1: 0.9498 - iou: 0.8743 - loss: 0.0102 - val_acc: 0.9013 - val_f1: 0.9060 - val_iou: 0.7882 - val_loss: 0.0183
Epoch 78/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9514 - f1: 0.9519 - iou: 0.8800 - loss: 0.0106 - val_acc: 0.9171 - val_f1: 0.9201 - val_iou: 0.8140 - val_loss: 0.0160
Epoch 79/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9480 - f1: 0.9488 - iou: 0.8691 - loss: 0.0110 - val_acc: 0.9244 - val_f1: 0.9265 - val_iou: 0.8258 - val_loss: 0.0134
Epoch 80/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9382 - f1: 0.9394 - iou: 0.8458 - loss: 0.0116 - val_acc: 0.9140 - val_f1: 0.9174 - val_iou: 0.8089 - val_loss: 0.0148
Epoch 81/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9425 - f1: 0.9436 - iou: 0.8597 - loss: 0.0117 - val_acc: 0.9157 - val_f1: 0.9189 - val_iou: 0.8120 - val_loss: 0.0161
Epoch 82/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9513 - f1: 0.9517 - iou: 0.8822 - loss: 0.0109 - val_acc: 0.9190 - val_f1: 0.9219 - val_iou: 0.8177 - val_loss: 0.0141
Epoch 83/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9527 - f1: 0.9534 - iou: 0.8803 - loss: 0.0101 - val_acc: 0.9193 - val_f1: 0.9221 - val_iou: 0.8178 - val_loss: 0.0154
Epoch 84/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9502 - f1: 0.9508 - iou: 0.8778 - loss: 0.0110 - val_acc: 0.9191 - val_f1: 0.9220 - val_iou: 0.8176 - val_loss: 0.0142
Epoch 85/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9410 - f1: 0.9419 - iou: 0.8604 - loss: 0.0115 - val_acc: 0.9278 - val_f1: 0.9297 - val_iou: 0.8324 - val_loss: 0.0130
Epoch 86/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9442 - f1: 0.9453 - iou: 0.8597 - loss: 0.0108 - val_acc: 0.9228 - val_f1: 0.9254 - val_iou: 0.8243 - val_loss: 0.0144
Epoch 87/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9446 - f1: 0.9455 - iou: 0.8676 - loss: 0.0108 - val_acc: 0.9276 - val_f1: 0.9297 - val_iou: 0.8328 - val_loss: 0.0131
Epoch 88/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9504 - f1: 0.9511 - iou: 0.8790 - loss: 0.0101 - val_acc: 0.9083 - val_f1: 0.9124 - val_iou: 0.8000 - val_loss: 0.0184
Epoch 89/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9469 - f1: 0.9479 - iou: 0.8692 - loss: 0.0106 - val_acc: 0.9253 - val_f1: 0.9277 - val_iou: 0.8287 - val_loss: 0.0135
Epoch 90/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9455 - f1: 0.9463 - iou: 0.8681 - loss: 0.0108 - val_acc: 0.9295 - val_f1: 0.9318 - val_iou: 0.8378 - val_loss: 0.0135
Epoch 91/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9455 - f1: 0.9462 - iou: 0.8664 - loss: 0.0104 - val_acc: 0.9224 - val_f1: 0.9251 - val_iou: 0.8239 - val_loss: 0.0137
Epoch 92/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9446 - f1: 0.9451 - iou: 0.8661 - loss: 0.0110 - val_acc: 0.9234 - val_f1: 0.9261 - val_iou: 0.8263 - val_loss: 0.0148
Epoch 93/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9420 - f1: 0.9431 - iou: 0.8610 - loss: 0.0116 - val_acc: 0.9197 - val_f1: 0.9227 - val_iou: 0.8193 - val_loss: 0.0143
Epoch 94/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 19ms/step - acc: 0.9508 - f1: 0.9513 - iou: 0.8832 - loss: 0.0104 - val_acc: 0.9196 - val_f1: 0.9225 - val_iou: 0.8190 - val_loss: 0.0139
Epoch 95/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9502 - f1: 0.9506 - iou: 0.8771 - loss: 0.0102 - val_acc: 0.9295 - val_f1: 0.9314 - val_iou: 0.8361 - val_loss: 0.0131
Epoch 96/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9570 - f1: 0.9574 - iou: 0.8918 - loss: 0.0092 - val_acc: 0.9348 - val_f1: 0.9358 - val_iou: 0.8438 - val_loss: 0.0123
Epoch 97/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 18ms/step - acc: 0.9497 - f1: 0.9502 - iou: 0.8796 - loss: 0.0106 - val_acc: 0.9120 - val_f1: 0.9157 - val_iou: 0.8062 - val_loss: 0.0152
Epoch 98/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9431 - f1: 0.9442 - iou: 0.8592 - loss: 0.0114 - val_acc: 0.9139 - val_f1: 0.9175 - val_iou: 0.8100 - val_loss: 0.0157
Epoch 99/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9483 - f1: 0.9489 - iou: 0.8771 - loss: 0.0111 - val_acc: 0.9368 - val_f1: 0.9382 - val_iou: 0.8501 - val_loss: 0.0120
Epoch 100/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9413 - f1: 0.9421 - iou: 0.8579 - loss: 0.0110 - val_acc: 0.9081 - val_f1: 0.9123 - val_iou: 0.8000 - val_loss: 0.0158
Epoch 101/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9489 - f1: 0.9498 - iou: 0.8752 - loss: 0.0100 - val_acc: 0.9314 - val_f1: 0.9333 - val_iou: 0.8402 - val_loss: 0.0124
Epoch 102/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9418 - f1: 0.9427 - iou: 0.8575 - loss: 0.0111 - val_acc: 0.9379 - val_f1: 0.9388 - val_iou: 0.8503 - val_loss: 0.0118
Epoch 103/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9516 - f1: 0.9522 - iou: 0.8802 - loss: 0.0101 - val_acc: 0.9332 - val_f1: 0.9349 - val_iou: 0.8433 - val_loss: 0.0130
Epoch 104/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9458 - f1: 0.9468 - iou: 0.8600 - loss: 0.0101 - val_acc: 0.9303 - val_f1: 0.9323 - val_iou: 0.8381 - val_loss: 0.0126
Epoch 105/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 33ms/step - acc: 0.9438 - f1: 0.9448 - iou: 0.8622 - loss: 0.0108 - val_acc: 0.9275 - val_f1: 0.9297 - val_iou: 0.8331 - val_loss: 0.0125
Epoch 106/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - acc: 0.9514 - f1: 0.9520 - iou: 0.8796 - loss: 0.0106 - val_acc: 0.9162 - val_f1: 0.9195 - val_iou: 0.8134 - val_loss: 0.0147
Epoch 107/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9446 - f1: 0.9457 - iou: 0.8631 - loss: 0.0107 - val_acc: 0.9249 - val_f1: 0.9275 - val_iou: 0.8290 - val_loss: 0.0138
Epoch 108/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 36ms/step - acc: 0.9493 - f1: 0.9502 - iou: 0.8735 - loss: 0.0100 - val_acc: 0.9346 - val_f1: 0.9364 - val_iou: 0.8466 - val_loss: 0.0119
Epoch 109/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9518 - f1: 0.9525 - iou: 0.8761 - loss: 0.0102 - val_acc: 0.9309 - val_f1: 0.9331 - val_iou: 0.8405 - val_loss: 0.0128
Epoch 110/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9451 - f1: 0.9464 - iou: 0.8617 - loss: 0.0105 - val_acc: 0.9307 - val_f1: 0.9327 - val_iou: 0.8394 - val_loss: 0.0124
Epoch 111/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - acc: 0.9495 - f1: 0.9501 - iou: 0.8743 - loss: 0.0099 - val_acc: 0.9312 - val_f1: 0.9332 - val_iou: 0.8401 - val_loss: 0.0125
Epoch 112/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9540 - f1: 0.9547 - iou: 0.8830 - loss: 0.0095 - val_acc: 0.9202 - val_f1: 0.9233 - val_iou: 0.8210 - val_loss: 0.0150
Epoch 113/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 36ms/step - acc: 0.9475 - f1: 0.9487 - iou: 0.8701 - loss: 0.0103 - val_acc: 0.9338 - val_f1: 0.9355 - val_iou: 0.8448 - val_loss: 0.0119
Epoch 114/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9517 - f1: 0.9525 - iou: 0.8793 - loss: 0.0099 - val_acc: 0.9360 - val_f1: 0.9375 - val_iou: 0.8485 - val_loss: 0.0123
Epoch 115/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9447 - f1: 0.9454 - iou: 0.8623 - loss: 0.0113 - val_acc: 0.9244 - val_f1: 0.9270 - val_iou: 0.8280 - val_loss: 0.0129
Epoch 116/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9503 - f1: 0.9509 - iou: 0.8784 - loss: 0.0100 - val_acc: 0.9249 - val_f1: 0.9273 - val_iou: 0.8280 - val_loss: 0.0136
Epoch 117/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9520 - f1: 0.9525 - iou: 0.8810 - loss: 0.0101 - val_acc: 0.9227 - val_f1: 0.9255 - val_iou: 0.8253 - val_loss: 0.0136
Epoch 118/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - acc: 0.9516 - f1: 0.9522 - iou: 0.8837 - loss: 0.0102 - val_acc: 0.9322 - val_f1: 0.9343 - val_iou: 0.8431 - val_loss: 0.0138
Epoch 119/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9521 - f1: 0.9528 - iou: 0.8843 - loss: 0.0099 - val_acc: 0.9248 - val_f1: 0.9275 - val_iou: 0.8294 - val_loss: 0.0147
Epoch 120/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - acc: 0.9550 - f1: 0.9556 - iou: 0.8865 - loss: 0.0094 - val_acc: 0.9344 - val_f1: 0.9362 - val_iou: 0.8464 - val_loss: 0.0120
Epoch 121/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9455 - f1: 0.9464 - iou: 0.8658 - loss: 0.0108 - val_acc: 0.9359 - val_f1: 0.9376 - val_iou: 0.8496 - val_loss: 0.0120
Epoch 122/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9517 - f1: 0.9525 - iou: 0.8791 - loss: 0.0100 - val_acc: 0.9236 - val_f1: 0.9263 - val_iou: 0.8268 - val_loss: 0.0130
Epoch 123/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9565 - f1: 0.9571 - iou: 0.8947 - loss: 0.0092 - val_acc: 0.9259 - val_f1: 0.9286 - val_iou: 0.8316 - val_loss: 0.0146
Epoch 124/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9477 - f1: 0.9487 - iou: 0.8694 - loss: 0.0102 - val_acc: 0.9288 - val_f1: 0.9313 - val_iou: 0.8370 - val_loss: 0.0129
Epoch 125/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 0s 20ms/step - acc: 0.9469 - f1: 0.9480 - iou: 0.8663 - loss: 0.0102 - val_acc: 0.9231 - val_f1: 0.9260 - val_iou: 0.8265 - val_loss: 0.0134
Epoch 126/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9491 - f1: 0.9499 - iou: 0.8731 - loss: 0.0097 - val_acc: 0.9270 - val_f1: 0.9295 - val_iou: 0.8335 - val_loss: 0.0132
Epoch 127/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 37ms/step - acc: 0.9496 - f1: 0.9505 - iou: 0.8692 - loss: 0.0097 - val_acc: 0.9392 - val_f1: 0.9406 - val_iou: 0.8553 - val_loss: 0.0113
Epoch 128/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9505 - f1: 0.9510 - iou: 0.8845 - loss: 0.0100 - val_acc: 0.9387 - val_f1: 0.9401 - val_iou: 0.8542 - val_loss: 0.0114
Epoch 129/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9448 - f1: 0.9453 - iou: 0.8596 - loss: 0.0109 - val_acc: 0.9356 - val_f1: 0.9375 - val_iou: 0.8495 - val_loss: 0.0119
Epoch 130/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9479 - f1: 0.9485 - iou: 0.8739 - loss: 0.0107 - val_acc: 0.9318 - val_f1: 0.9340 - val_iou: 0.8425 - val_loss: 0.0128
Epoch 131/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9580 - f1: 0.9584 - iou: 0.8994 - loss: 0.0091 - val_acc: 0.9331 - val_f1: 0.9351 - val_iou: 0.8444 - val_loss: 0.0126
Epoch 132/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9563 - f1: 0.9568 - iou: 0.8925 - loss: 0.0093 - val_acc: 0.9254 - val_f1: 0.9280 - val_iou: 0.8300 - val_loss: 0.0130
Epoch 133/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9492 - f1: 0.9501 - iou: 0.8690 - loss: 0.0105 - val_acc: 0.9268 - val_f1: 0.9293 - val_iou: 0.8324 - val_loss: 0.0125
Epoch 134/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9517 - f1: 0.9527 - iou: 0.8806 - loss: 0.0095 - val_acc: 0.9232 - val_f1: 0.9261 - val_iou: 0.8265 - val_loss: 0.0132
Epoch 135/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9548 - f1: 0.9553 - iou: 0.8878 - loss: 0.0092 - val_acc: 0.9273 - val_f1: 0.9297 - val_iou: 0.8335 - val_loss: 0.0125
Epoch 136/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - acc: 0.9474 - f1: 0.9478 - iou: 0.8689 - loss: 0.0102 - val_acc: 0.9279 - val_f1: 0.9304 - val_iou: 0.8349 - val_loss: 0.0137
Epoch 137/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9513 - f1: 0.9523 - iou: 0.8797 - loss: 0.0101 - val_acc: 0.9353 - val_f1: 0.9371 - val_iou: 0.8487 - val_loss: 0.0124
Epoch 138/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9526 - f1: 0.9531 - iou: 0.8869 - loss: 0.0100 - val_acc: 0.9325 - val_f1: 0.9346 - val_iou: 0.8436 - val_loss: 0.0122
Epoch 139/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9512 - f1: 0.9518 - iou: 0.8831 - loss: 0.0093 - val_acc: 0.9273 - val_f1: 0.9298 - val_iou: 0.8340 - val_loss: 0.0138
Epoch 140/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9534 - f1: 0.9541 - iou: 0.8865 - loss: 0.0091 - val_acc: 0.9411 - val_f1: 0.9422 - val_iou: 0.8585 - val_loss: 0.0112
Epoch 141/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9524 - f1: 0.9529 - iou: 0.8814 - loss: 0.0096 - val_acc: 0.9325 - val_f1: 0.9346 - val_iou: 0.8437 - val_loss: 0.0123
Epoch 142/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - acc: 0.9473 - f1: 0.9480 - iou: 0.8716 - loss: 0.0101 - val_acc: 0.9386 - val_f1: 0.9402 - val_iou: 0.8547 - val_loss: 0.0113
Epoch 143/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9483 - f1: 0.9490 - iou: 0.8731 - loss: 0.0100 - val_acc: 0.9389 - val_f1: 0.9405 - val_iou: 0.8557 - val_loss: 0.0112
Epoch 144/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9517 - f1: 0.9523 - iou: 0.8841 - loss: 0.0096 - val_acc: 0.9267 - val_f1: 0.9293 - val_iou: 0.8328 - val_loss: 0.0127
Epoch 145/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9516 - f1: 0.9524 - iou: 0.8816 - loss: 0.0097 - val_acc: 0.9384 - val_f1: 0.9400 - val_iou: 0.8544 - val_loss: 0.0114
Epoch 146/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9535 - f1: 0.9537 - iou: 0.8898 - loss: 0.0095 - val_acc: 0.9317 - val_f1: 0.9338 - val_iou: 0.8420 - val_loss: 0.0128
Epoch 147/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9564 - f1: 0.9569 - iou: 0.8933 - loss: 0.0088 - val_acc: 0.9394 - val_f1: 0.9409 - val_iou: 0.8562 - val_loss: 0.0117
Epoch 148/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9532 - f1: 0.9538 - iou: 0.8774 - loss: 0.0090 - val_acc: 0.9281 - val_f1: 0.9306 - val_iou: 0.8356 - val_loss: 0.0131
Epoch 149/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9555 - f1: 0.9559 - iou: 0.8915 - loss: 0.0094 - val_acc: 0.9353 - val_f1: 0.9373 - val_iou: 0.8494 - val_loss: 0.0126
Epoch 150/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9530 - f1: 0.9537 - iou: 0.8853 - loss: 0.0095 - val_acc: 0.9207 - val_f1: 0.9237 - val_iou: 0.8220 - val_loss: 0.0143
Epoch 151/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9575 - f1: 0.9582 - iou: 0.8921 - loss: 0.0087 - val_acc: 0.9392 - val_f1: 0.9407 - val_iou: 0.8559 - val_loss: 0.0113
Epoch 152/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9489 - f1: 0.9495 - iou: 0.8713 - loss: 0.0099 - val_acc: 0.9330 - val_f1: 0.9350 - val_iou: 0.8445 - val_loss: 0.0122
Epoch 153/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9499 - f1: 0.9503 - iou: 0.8806 - loss: 0.0096 - val_acc: 0.9405 - val_f1: 0.9419 - val_iou: 0.8585 - val_loss: 0.0113
Epoch 154/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9460 - f1: 0.9468 - iou: 0.8681 - loss: 0.0100 - val_acc: 0.9386 - val_f1: 0.9402 - val_iou: 0.8549 - val_loss: 0.0113
Epoch 155/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9533 - f1: 0.9535 - iou: 0.8871 - loss: 0.0094 - val_acc: 0.9362 - val_f1: 0.9381 - val_iou: 0.8508 - val_loss: 0.0120
Epoch 156/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9489 - f1: 0.9498 - iou: 0.8694 - loss: 0.0100 - val_acc: 0.9375 - val_f1: 0.9394 - val_iou: 0.8537 - val_loss: 0.0117
Epoch 157/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9540 - f1: 0.9544 - iou: 0.8870 - loss: 0.0090 - val_acc: 0.9296 - val_f1: 0.9320 - val_iou: 0.8385 - val_loss: 0.0130
Epoch 158/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9604 - f1: 0.9608 - iou: 0.8996 - loss: 0.0084 - val_acc: 0.9320 - val_f1: 0.9342 - val_iou: 0.8427 - val_loss: 0.0128
Epoch 159/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9566 - f1: 0.9569 - iou: 0.8928 - loss: 0.0087 - val_acc: 0.9344 - val_f1: 0.9364 - val_iou: 0.8475 - val_loss: 0.0124
Epoch 160/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9624 - f1: 0.9630 - iou: 0.9030 - loss: 0.0082 - val_acc: 0.9346 - val_f1: 0.9366 - val_iou: 0.8477 - val_loss: 0.0119
Epoch 161/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9576 - f1: 0.9580 - iou: 0.8941 - loss: 0.0089 - val_acc: 0.9300 - val_f1: 0.9323 - val_iou: 0.8393 - val_loss: 0.0134
Epoch 162/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9606 - f1: 0.9612 - iou: 0.9038 - loss: 0.0085 - val_acc: 0.9340 - val_f1: 0.9361 - val_iou: 0.8469 - val_loss: 0.0127
Epoch 163/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9546 - f1: 0.9551 - iou: 0.8882 - loss: 0.0095 - val_acc: 0.9420 - val_f1: 0.9433 - val_iou: 0.8615 - val_loss: 0.0111
Epoch 164/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9460 - f1: 0.9470 - iou: 0.8642 - loss: 0.0100 - val_acc: 0.9402 - val_f1: 0.9417 - val_iou: 0.8582 - val_loss: 0.0112
Epoch 165/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9490 - f1: 0.9497 - iou: 0.8757 - loss: 0.0098 - val_acc: 0.9316 - val_f1: 0.9338 - val_iou: 0.8420 - val_loss: 0.0121
Epoch 166/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9565 - f1: 0.9568 - iou: 0.8938 - loss: 0.0088 - val_acc: 0.9299 - val_f1: 0.9323 - val_iou: 0.8392 - val_loss: 0.0129
Epoch 167/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9528 - f1: 0.9534 - iou: 0.8868 - loss: 0.0095 - val_acc: 0.9340 - val_f1: 0.9361 - val_iou: 0.8471 - val_loss: 0.0124
Epoch 168/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9547 - f1: 0.9554 - iou: 0.8871 - loss: 0.0095 - val_acc: 0.9379 - val_f1: 0.9397 - val_iou: 0.8543 - val_loss: 0.0117
Epoch 169/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9583 - f1: 0.9587 - iou: 0.8958 - loss: 0.0090 - val_acc: 0.9330 - val_f1: 0.9351 - val_iou: 0.8449 - val_loss: 0.0126
Epoch 170/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - acc: 0.9584 - f1: 0.9588 - iou: 0.8966 - loss: 0.0089 - val_acc: 0.9307 - val_f1: 0.9330 - val_iou: 0.8406 - val_loss: 0.0129
Epoch 171/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - acc: 0.9616 - f1: 0.9620 - iou: 0.9034 - loss: 0.0083 - val_acc: 0.9360 - val_f1: 0.9380 - val_iou: 0.8508 - val_loss: 0.0123
Epoch 172/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 33ms/step - acc: 0.9563 - f1: 0.9565 - iou: 0.8916 - loss: 0.0092 - val_acc: 0.9326 - val_f1: 0.9348 - val_iou: 0.8443 - val_loss: 0.0129
Epoch 173/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9524 - f1: 0.9530 - iou: 0.8802 - loss: 0.0092 - val_acc: 0.9340 - val_f1: 0.9361 - val_iou: 0.8469 - val_loss: 0.0123
Epoch 174/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - acc: 0.9523 - f1: 0.9530 - iou: 0.8803 - loss: 0.0095 - val_acc: 0.9386 - val_f1: 0.9404 - val_iou: 0.8557 - val_loss: 0.0116
Epoch 175/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - acc: 0.9529 - f1: 0.9536 - iou: 0.8810 - loss: 0.0091 - val_acc: 0.9372 - val_f1: 0.9390 - val_iou: 0.8529 - val_loss: 0.0118
Epoch 176/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 35ms/step - acc: 0.9571 - f1: 0.9574 - iou: 0.8940 - loss: 0.0089 - val_acc: 0.9342 - val_f1: 0.9362 - val_iou: 0.8471 - val_loss: 0.0124
Epoch 177/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9430 - f1: 0.9437 - iou: 0.8637 - loss: 0.0106 - val_acc: 0.9329 - val_f1: 0.9350 - val_iou: 0.8446 - val_loss: 0.0124
Epoch 178/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9535 - f1: 0.9541 - iou: 0.8860 - loss: 0.0094 - val_acc: 0.9324 - val_f1: 0.9346 - val_iou: 0.8438 - val_loss: 0.0124
Epoch 179/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9615 - f1: 0.9619 - iou: 0.9038 - loss: 0.0082 - val_acc: 0.9324 - val_f1: 0.9346 - val_iou: 0.8437 - val_loss: 0.0125
Epoch 180/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9557 - f1: 0.9561 - iou: 0.8909 - loss: 0.0091 - val_acc: 0.9328 - val_f1: 0.9349 - val_iou: 0.8445 - val_loss: 0.0124
Epoch 181/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9544 - f1: 0.9547 - iou: 0.8889 - loss: 0.0096 - val_acc: 0.9324 - val_f1: 0.9346 - val_iou: 0.8437 - val_loss: 0.0125
Epoch 182/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 21ms/step - acc: 0.9514 - f1: 0.9521 - iou: 0.8815 - loss: 0.0096 - val_acc: 0.9325 - val_f1: 0.9347 - val_iou: 0.8439 - val_loss: 0.0124
Epoch 183/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 22ms/step - acc: 0.9459 - f1: 0.9462 - iou: 0.8747 - loss: 0.0115 - val_acc: 0.9330 - val_f1: 0.9351 - val_iou: 0.8449 - val_loss: 0.0123
Epoch 184/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9504 - f1: 0.9509 - iou: 0.8806 - loss: 0.0093 - val_acc: 0.9352 - val_f1: 0.9372 - val_iou: 0.8490 - val_loss: 0.0119
Epoch 185/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9499 - f1: 0.9503 - iou: 0.8773 - loss: 0.0093 - val_acc: 0.9353 - val_f1: 0.9372 - val_iou: 0.8491 - val_loss: 0.0119
Epoch 186/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - acc: 0.9576 - f1: 0.9583 - iou: 0.8935 - loss: 0.0091 - val_acc: 0.9352 - val_f1: 0.9372 - val_iou: 0.8490 - val_loss: 0.0119
Epoch 187/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9477 - f1: 0.9484 - iou: 0.8735 - loss: 0.0097 - val_acc: 0.9350 - val_f1: 0.9369 - val_iou: 0.8485 - val_loss: 0.0119
Epoch 188/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 27ms/step - acc: 0.9592 - f1: 0.9597 - iou: 0.8980 - loss: 0.0085 - val_acc: 0.9346 - val_f1: 0.9366 - val_iou: 0.8478 - val_loss: 0.0120
Epoch 189/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9576 - f1: 0.9579 - iou: 0.8971 - loss: 0.0087 - val_acc: 0.9343 - val_f1: 0.9364 - val_iou: 0.8473 - val_loss: 0.0120
Epoch 190/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 26ms/step - acc: 0.9539 - f1: 0.9544 - iou: 0.8848 - loss: 0.0085 - val_acc: 0.9340 - val_f1: 0.9360 - val_iou: 0.8466 - val_loss: 0.0121
Epoch 191/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9629 - f1: 0.9633 - iou: 0.9051 - loss: 0.0077 - val_acc: 0.9338 - val_f1: 0.9359 - val_iou: 0.8464 - val_loss: 0.0122
Epoch 192/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 29ms/step - acc: 0.9532 - f1: 0.9538 - iou: 0.8860 - loss: 0.0097 - val_acc: 0.9340 - val_f1: 0.9360 - val_iou: 0.8467 - val_loss: 0.0122
Epoch 193/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 31ms/step - acc: 0.9605 - f1: 0.9610 - iou: 0.8996 - loss: 0.0084 - val_acc: 0.9341 - val_f1: 0.9361 - val_iou: 0.8468 - val_loss: 0.0122
Epoch 194/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 30ms/step - acc: 0.9522 - f1: 0.9528 - iou: 0.8815 - loss: 0.0095 - val_acc: 0.9340 - val_f1: 0.9361 - val_iou: 0.8468 - val_loss: 0.0122
Epoch 195/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 32ms/step - acc: 0.9529 - f1: 0.9535 - iou: 0.8859 - loss: 0.0093 - val_acc: 0.9339 - val_f1: 0.9359 - val_iou: 0.8465 - val_loss: 0.0122
Epoch 196/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 24ms/step - acc: 0.9525 - f1: 0.9534 - iou: 0.8785 - loss: 0.0095 - val_acc: 0.9337 - val_f1: 0.9358 - val_iou: 0.8462 - val_loss: 0.0122
Epoch 197/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 25ms/step - acc: 0.9536 - f1: 0.9542 - iou: 0.8846 - loss: 0.0089 - val_acc: 0.9337 - val_f1: 0.9358 - val_iou: 0.8461 - val_loss: 0.0122
Epoch 198/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 23ms/step - acc: 0.9636 - f1: 0.9640 - iou: 0.9078 - loss: 0.0079 - val_acc: 0.9334 - val_f1: 0.9355 - val_iou: 0.8457 - val_loss: 0.0123
Epoch 199/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 33ms/step - acc: 0.9538 - f1: 0.9544 - iou: 0.8844 - loss: 0.0097 - val_acc: 0.9335 - val_f1: 0.9356 - val_iou: 0.8458 - val_loss: 0.0123
Epoch 200/200
25/25 ━━━━━━━━━━━━━━━━━━━━ 1s 28ms/step - acc: 0.9491 - f1: 0.9499 - iou: 0.8771 - loss: 0.0099 - val_acc: 0.9336 - val_f1: 0.9357 - val_iou: 0.8459 - val_loss: 0.0123
31/31 ━━━━━━━━━━━━━━━━━━━━ 1s 918us/step
31/31 ━━━━━━━━━━━━━━━━━━━━ 0s 1ms/step - acc: 0.9448 - f1: 0.9461 - iou: 0.8636 - loss: 0.0111

INFO     [TEST SET] acc=94.20%, f1=94.33%, iou=86.15%, loss=1.11%                                      train.py:221

Model evaluation¶

Now that we have trained the model, we will evaluate the model on the test dataset. Similar to training, we will provide the high-level configuration to the task process.

In [22]:

task.evaluate(params)

task.evaluate(params)

Subject: 100%|██████████| 56/56 [00:17<00:00,  3.21it/s]

INFO     Testing Results                                                                            evaluate.py:130

19/19 ━━━━━━━━━━━━━━━━━━━━ 2s 25ms/step - acc: 0.9436 - f1: 0.9448 - iou: 0.8600 - loss: 0.0101

INFO     [TEST SET] acc=94.59%, f1=94.65%, iou=86.69%, loss=0.97%                                   evaluate.py:132

Confusion matrix¶

Let's visualize the confusion matrix to understand the model's performance on each class.

In [23]:

IPython.display.Image(filename=params.job_dir / "confusion_matrix_test.png", width=500)

IPython.display.Image(filename=params.job_dir / "confusion_matrix_test.png", width=500)

Out[23]:

Export model to TF Lite / TFLM¶

Once we have trained and evaluated the model, we need to export the model into a format that can be used for inference on the edge. Currently, we export the model to TensorFlow Lite flatbuffer format. This will also generate a C header file that can be used with TensorFlow Lite for Microcontrollers (TFLM).

Apply post-training quantization (PTQ)¶

For running on bare metal, we will perform post-training quantization to convert the model to an 8-bit integer model. The weights and activations will be quantized to 8-bits and biases will be quantized to 32-bits. This will reduce the model size and improve the inference speed.

In [28]:

quantization = sk.QuantizationParams(
    enabled=True,
    format="FP32",
    io_type="float32",
    conversion="CONCRETE",
)
params.quantization = quantization

quantization = sk.QuantizationParams( enabled=True, format="FP32", io_type="float32", conversion="CONCRETE", ) params.quantization = quantization

In [29]:

# TF dumps a lot of info to stdout, so we redirect it to /dev/null
with open(os.devnull, 'w') as devnull:
    with contextlib.redirect_stdout(devnull), contextlib.redirect_stderr(devnull):
        task.export(params)

# TF dumps a lot of info to stdout, so we redirect it to /dev/null with open(os.devnull, 'w') as devnull: with contextlib.redirect_stdout(devnull), contextlib.redirect_stderr(devnull): task.export(params)

[10/02/24 15:01:29] WARNING  WARNING:absl:Please consider providing the trackable_obj argument in the  lite.py:2166
                             from_concrete_functions. Providing without the trackable_obj argument is              
                             deprecated and it will use the deprecated conversion path.                            

I0000 00:00:1727881289.426610    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727881289.426701    5610 devices.cc:67] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 1
I0000 00:00:1727881289.427005    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727881289.427061    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727881289.427104    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727881289.427172    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727881289.427216    5610 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
W0000 00:00:1727881289.517520    5610 tf_tfl_flatbuffer_helpers.cc:392] Ignored output_format.
W0000 00:00:1727881289.517532    5610 tf_tfl_flatbuffer_helpers.cc:395] Ignored drop_control_dependency.

INFO     [TF METRICS] LOSS=0.1925 ACC=0.9566 F1=0.9568                                                export.py:111

INFO     [TFL METRICS] LOSS=0.1925 ACC=0.9566 F1=0.9568                                               export.py:112

INFO     Validation passed (0.00%)                                                                    export.py:120

Run inference demo¶

We will run a demo on the PC to verify that the model is working as expected. The demo will load the model and run inferences across a randomly selected subject. The demo will also provide the model's prediction and the corresponding class name.

In [9]:

task.demo(params=params)

task.demo(params=params)

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1727883197.048706  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.068145  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.068255  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.069518  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.069592  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.069636  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.117509  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.117605  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1727883197.117663  223691 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
Inference:   0%|          | 0/296 [00:00<?, ?it/s]WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1727883197.752020  223865 service.cc:146] XLA service 0x741ed800fff0 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1727883197.752040  223865 service.cc:154]   StreamExecutor device (0): NVIDIA GeForce RTX 4090, Compute Capability 8.9
I0000 00:00:1727883198.494078  223865 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.
Inference: 100%|██████████| 296/296 [00:06<00:00, 42.97it/s]