Spaces:
Sleeping
Sleeping
File size: 10,156 Bytes
e4aef33 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 |
import os
import shutil
import gradio as gr
import pandas as pd
from utils import run_supervisely_parser, train_model
def run_pipeline(
base_model_zip,
supervisely_project_zip,
train_ratio,
seed,
data_percent,
batch_size,
num_epochs,
learning_rate,
image_width,
image_height,
early_stopping,
validate_every,
pr=gr.Progress(track_tqdm=True),
):
# Parsing
yield (
gr.update(interactive=False), # run button
gr.update( # status textbox
value="Parsing Supervisely project ...",
visible=True,
),
gr.update(visible=False), # model download button
gr.update(visible=False), # output tab
gr.update(value=None), # train IoU plot
gr.update(value=None), # val IoU plot
gr.update(value=None), # metrics table
)
dataset_dir = run_supervisely_parser(
project_path=supervisely_project_zip,
train_ratio=train_ratio,
seed=seed,
)
# Training
yield (
gr.update(interactive=False), # run button
gr.update( # status textbox
value="Starting model training...",
),
gr.update(visible=False), # model download button
gr.update(visible=False), # output tab
gr.update(value=None), # train IoU plot
gr.update(value=None), # val IoU plot
gr.update(value=None), # metrics table
)
best_model, metrics, dice = train_model(
dataset_dir,
base_model_zip,
image_width,
image_height,
batch_size,
data_percent,
num_epochs,
learning_rate,
early_stopping,
validate_every,
)
# Saving model
yield (
gr.update(interactive=False), # run button
gr.update( # status textbox
value="Saving best model...",
),
gr.update(visible=False), # model download button
gr.update(visible=False), # output tab
gr.update(value=None), # train IoU plot
gr.update(value=None), # val IoU plot
gr.update(value=None), # metrics table
)
best_model_dir = os.path.join(
os.path.dirname(base_model_zip),
"best_model",
)
best_model.save_pretrained(best_model_dir)
best_model_zip_path = shutil.make_archive(
base_name=best_model_dir,
format="zip",
root_dir=best_model_dir,
)
metrics_df = pd.DataFrame(metrics)
initial_epoch_metrics = metrics_df.iloc[0]
final_epoch_metrics = metrics_df.iloc[-1]
# metrics comparison table use epoch 0 as before and final as after
metrics_comparison_df = pd.DataFrame(
{
"Metric": ["Accuracy", "IoU", "Loss", "Dice"],
"Before": [
initial_epoch_metrics["val_acc"],
initial_epoch_metrics["val_iou"],
initial_epoch_metrics["val_loss"],
dice[0],
],
"After": [
final_epoch_metrics["val_acc"],
final_epoch_metrics["val_iou"],
final_epoch_metrics["val_loss"],
dice[1],
],
}
)
yield (
gr.update(interactive=True), # run button
gr.update(visible=False), # status textbox
gr.update( # model download button
value=best_model_zip_path,
visible=True,
),
gr.update(visible=True), # output tab
gr.update(value=metrics_df), # train IoU plot
gr.update(value=metrics_df), # val IoU plot
gr.update(
value=metrics_comparison_df,
visible=True,
),
)
def _toggle_run_btn(base_model, project):
"""Enable run button only when both required files are selected."""
ready = bool(base_model and project)
return gr.update(interactive=ready)
with gr.Blocks(title="SegFormer Training & Dataset Pipeline") as demo:
gr.Markdown(
"# SegFormer Training Pipeline\n"
"Upload your base model and Supervisely project, "
"tweak parsing & training hyperparameters, then click "
"**Run Training**."
)
with gr.Row():
base_model_zip = gr.File(
label="Base PyTorch Model (.zip)",
file_types=[".zip"],
file_count="single",
)
supervisely_project_zip = gr.File(
label="Supervisely Project (.zip)",
file_types=[".zip"],
file_count="single",
)
with gr.Tab("Training"):
gr.Markdown("Adjust training hyperparameters.")
with gr.Row():
data_percent = gr.Slider(
minimum=1,
maximum=100,
step=1,
value=100,
label="Data Percent (%) used for training",
)
batch_size = gr.Number(
value=32,
label="Batch Size (samples/step)",
precision=0,
minimum=1,
)
num_epochs = gr.Number(
value=60,
label="Epochs (max passes)",
precision=0,
minimum=1,
)
with gr.Row():
learning_rate = gr.Number(
value=5e-5,
label="Learning Rate",
minimum=0.0,
maximum=1.0,
)
image_width = gr.Number(
value=640,
label="Image Width (px)",
precision=0,
minimum=1,
)
image_height = gr.Number(
value=640,
label="Image Height (px)",
precision=0,
minimum=1,
)
with gr.Row():
early_stopping = gr.Number(
value=3,
label="Early Stopping Patience (epochs w/o improvement)",
precision=0,
minimum=0,
)
validate_every = gr.Number(
value=1,
label="Validate Every (epochs)",
precision=0,
minimum=0,
)
with gr.Tab("Dataset Parsing"):
gr.Markdown("Configure how the dataset is split and seeded.")
with gr.Row():
train_ratio = gr.Slider(
minimum=0.1,
maximum=0.95,
step=0.01,
value=0.8,
label="Train Split Ratio (rest used for validation)",
)
seed = gr.Number(
value=42,
label="Random Seed (reproducibility)",
precision=0,
)
with gr.Accordion("Parameter Help", open=False):
gr.Markdown(
"""
**Base PyTorch Model (.zip)**: Archive containing a folder with
weights and configuration file.\n
**Supervisely Project (.zip)**: Archive containing Exported
Supervisely project
containing images and annotation JSONs.\n
**Train Split Ratio**: Fraction of dataset used for training;
remainder becomes validation.\n
**Random Seed**: Controls shuffling for reproducible splits &
training.\n
**Data Percent**: Subsample percentage of training split (use
<100 for quick experiments).\n
**Batch Size**: Samples processed before each optimizer step.\n
**Epochs**: Maximum complete passes over the (subsampled)
training set.\n
**Learning Rate**: Initial optimizer step size.\n
**Image Width / Height**: Target spatial size for preprocessing
(resize/crop).\n
**Early Stopping Patience**: Stop after this many validation
checks without improvement.\n
**Validate Every**: Run validation after this many epochs.\n
"""
)
run_btn = gr.Button(
"Run Training",
variant="primary",
interactive=False,
)
status = gr.Textbox(
show_label=False,
visible=False,
)
with gr.Tab("Results", visible=False) as output_tab:
model_download_btn = gr.DownloadButton(
label="Download Trained Model (.zip)",
value=None,
visible=False,
)
# table to show before and after accuracy and iou
metrics_table = gr.DataFrame(
label="Metrics Comparison",
interactive=False,
wrap=True,
)
with gr.Row():
train_iou_plot = gr.LinePlot(
label="Training IoU",
x="epoch",
y="train_iou",
x_title="Epoch",
y_title="IoU",
height=400,
)
val_iou_plot = gr.LinePlot(
label="Validation IoU",
x="epoch",
y="val_iou",
x_title="Epoch",
y_title="IoU",
height=400,
)
# Enable run button only when both archives provided
base_model_zip.change(
_toggle_run_btn,
inputs=[base_model_zip, supervisely_project_zip],
outputs=run_btn,
)
supervisely_project_zip.change(
_toggle_run_btn,
inputs=[base_model_zip, supervisely_project_zip],
outputs=run_btn,
)
# Click handler
run_btn.click(
run_pipeline,
inputs=[
base_model_zip,
supervisely_project_zip,
train_ratio,
seed,
data_percent,
batch_size,
num_epochs,
learning_rate,
image_width,
image_height,
early_stopping,
validate_every,
],
outputs=[
run_btn,
status,
model_download_btn,
output_tab,
train_iou_plot,
val_iou_plot,
metrics_table,
],
show_progress_on=status,
scroll_to_output=True,
)
if __name__ == "__main__":
demo.launch()
|