Upload app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
import gradio as gr
|
2 |
-
import spaces
|
3 |
import onnxruntime as ort
|
4 |
import numpy as np
|
5 |
from PIL import Image, ImageDraw, ImageFont
|
@@ -12,6 +12,8 @@ import matplotlib
|
|
12 |
from huggingface_hub import hf_hub_download
|
13 |
from dataclasses import dataclass
|
14 |
from typing import List, Dict, Optional, Tuple
|
|
|
|
|
15 |
|
16 |
# MatplotlibのバックエンドをAggに設定 (GUIなし環境用)
|
17 |
matplotlib.use('Agg')
|
@@ -293,7 +295,8 @@ TAG_MAPPING_FILENAME = "cl_eva02_tagger_v1_250426/tag_mapping.json"
|
|
293 |
CACHE_DIR = "./model_cache"
|
294 |
|
295 |
# グローバル変数(モデルとラベルをキャッシュ)
|
296 |
-
onnx_session = None
|
|
|
297 |
labels_data = None
|
298 |
tag_to_category_map = None
|
299 |
|
@@ -319,111 +322,98 @@ def download_model_files():
|
|
319 |
|
320 |
|
321 |
def initialize_model():
|
322 |
-
"""
|
323 |
-
global
|
324 |
-
|
|
|
|
|
325 |
model_path, tag_mapping_path = download_model_files()
|
326 |
-
|
327 |
-
|
328 |
-
# --- Added Logging ---
|
329 |
-
print("--- Environment Check ---")
|
330 |
-
try:
|
331 |
-
import torch
|
332 |
-
print(f"PyTorch version: {torch.__version__}")
|
333 |
-
if torch.cuda.is_available():
|
334 |
-
print(f"PyTorch CUDA available: True")
|
335 |
-
print(f"PyTorch CUDA version: {torch.version.cuda}")
|
336 |
-
print(f"Detected GPU: {torch.cuda.get_device_name(0)}")
|
337 |
-
if torch.backends.cudnn.is_available():
|
338 |
-
print(f"PyTorch cuDNN available: True")
|
339 |
-
print(f"PyTorch cuDNN version: {torch.backends.cudnn.version()}")
|
340 |
-
else:
|
341 |
-
print("PyTorch cuDNN available: False")
|
342 |
-
else:
|
343 |
-
print("PyTorch CUDA available: False")
|
344 |
-
except ImportError:
|
345 |
-
print("PyTorch not found.")
|
346 |
-
except Exception as e:
|
347 |
-
print(f"Error during PyTorch check: {e}")
|
348 |
-
|
349 |
-
try:
|
350 |
-
print(f"ONNX Runtime build info: {ort.get_buildinfo()}")
|
351 |
-
except Exception as e:
|
352 |
-
print(f"Error getting ONNX Runtime build info: {e}")
|
353 |
-
print("-------------------------")
|
354 |
-
# --- End Added Logging ---
|
355 |
-
|
356 |
-
# ONNXセッションの初期化 (GPU優先)
|
357 |
-
available_providers = ort.get_available_providers()
|
358 |
-
print(f"Available ONNX Runtime providers: {available_providers}")
|
359 |
-
providers = []
|
360 |
-
if 'CUDAExecutionProvider' in available_providers:
|
361 |
-
providers.append('CUDAExecutionProvider')
|
362 |
-
# elif 'DmlExecutionProvider' in available_providers: # DirectML (Windows)
|
363 |
-
# providers.append('DmlExecutionProvider')
|
364 |
-
providers.append('CPUExecutionProvider') # Always include CPU as fallback
|
365 |
-
|
366 |
-
try:
|
367 |
-
onnx_session = ort.InferenceSession(model_path, providers=providers)
|
368 |
-
print(f"Using ONNX Runtime provider: {onnx_session.get_providers()[0]}")
|
369 |
-
except Exception as e:
|
370 |
-
print(f"Error initializing ONNX session with providers {providers}: {e}")
|
371 |
-
print("Falling back to CPUExecutionProvider only.")
|
372 |
-
onnx_session = ort.InferenceSession(model_path, providers=['CPUExecutionProvider'])
|
373 |
-
|
374 |
labels_data, _, tag_to_category_map = load_tag_mapping(tag_mapping_path)
|
375 |
-
print("
|
|
|
376 |
|
377 |
@spaces.GPU()
|
378 |
def predict(image_input, gen_threshold, char_threshold, output_mode):
|
379 |
-
print("--- predict function started ---")
|
380 |
-
"""Gradioインターフェース用の予測関数"""
|
381 |
-
initialize_model() #
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
382 |
|
383 |
if image_input is None:
|
384 |
return "Please upload an image.", None
|
385 |
|
386 |
-
print(f"Processing image with thresholds: gen={gen_threshold}, char={char_threshold}")
|
387 |
|
388 |
# PIL Imageオブジェクトであることを確認
|
389 |
if not isinstance(image_input, Image.Image):
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
|
394 |
-
|
395 |
-
|
396 |
-
|
397 |
-
|
398 |
-
|
399 |
-
|
400 |
-
|
401 |
-
|
402 |
-
|
403 |
-
|
404 |
-
|
405 |
-
|
406 |
-
|
407 |
else:
|
408 |
image = image_input
|
409 |
|
410 |
-
|
411 |
# 前処理
|
412 |
original_pil_image, input_data = preprocess_image(image)
|
413 |
|
414 |
# データ型をモデルの期待に合わせる (通常はfloat32)
|
415 |
-
input_name =
|
416 |
-
expected_type =
|
417 |
if expected_type == 'tensor(float16)':
|
418 |
input_data = input_data.astype(np.float16)
|
419 |
else:
|
420 |
input_data = input_data.astype(np.float32) # Default to float32
|
421 |
|
422 |
-
# 推論
|
423 |
start_time = time.time()
|
424 |
-
outputs =
|
425 |
inference_time = time.time() - start_time
|
426 |
-
print(f"Inference completed in {inference_time:.3f} seconds")
|
427 |
|
428 |
# シグモイド関数で確率に変換
|
429 |
probs = 1 / (1 + np.exp(-outputs[0])) # Apply sigmoid to the first batch item
|
@@ -437,12 +427,12 @@ def predict(image_input, gen_threshold, char_threshold, output_mode):
|
|
437 |
if predictions["rating"]:
|
438 |
output_tags.append(predictions["rating"][0][0].replace("_", " "))
|
439 |
if predictions["quality"]:
|
440 |
-
|
441 |
|
442 |
# 残りのカテゴリをアルファベット順に追加(オプション)
|
443 |
for category in ["artist", "character", "copyright", "general", "meta"]:
|
444 |
tags = [tag.replace("_", " ") for tag, prob in predictions[category]
|
445 |
-
|
446 |
output_tags.extend(tags)
|
447 |
|
448 |
output_text = ", ".join(output_tags)
|
@@ -454,7 +444,6 @@ def predict(image_input, gen_threshold, char_threshold, output_mode):
|
|
454 |
return output_text, viz_image
|
455 |
|
456 |
# --- Gradio Interface Definition ---
|
457 |
-
import time
|
458 |
|
459 |
# CSS for styling
|
460 |
css = """
|
@@ -594,6 +583,5 @@ if __name__ == "__main__":
|
|
594 |
# 環境変数HF_TOKENがない場合に警告(プライベートリポジトリ用)
|
595 |
if not os.environ.get("HF_TOKEN"):
|
596 |
print("Warning: HF_TOKEN environment variable not set. Downloads from private repositories may fail.")
|
597 |
-
#
|
598 |
-
initialize_model() # Removed startup initialization
|
599 |
demo.launch(share=True)
|
|
|
1 |
import gradio as gr
|
2 |
+
# import spaces # Removed
|
3 |
import onnxruntime as ort
|
4 |
import numpy as np
|
5 |
from PIL import Image, ImageDraw, ImageFont
|
|
|
12 |
from huggingface_hub import hf_hub_download
|
13 |
from dataclasses import dataclass
|
14 |
from typing import List, Dict, Optional, Tuple
|
15 |
+
import time
|
16 |
+
import spaces
|
17 |
|
18 |
# MatplotlibのバックエンドをAggに設定 (GUIなし環境用)
|
19 |
matplotlib.use('Agg')
|
|
|
295 |
CACHE_DIR = "./model_cache"
|
296 |
|
297 |
# グローバル変数(モデルとラベルをキャッシュ)
|
298 |
+
# onnx_session = None # Removed global session
|
299 |
+
model_path_global = None # Store model path globally
|
300 |
labels_data = None
|
301 |
tag_to_category_map = None
|
302 |
|
|
|
322 |
|
323 |
|
324 |
def initialize_model():
|
325 |
+
"""モデルファイルとラベルデータを準備(キャッシュ)"""
|
326 |
+
global model_path_global, labels_data, tag_to_category_map
|
327 |
+
# Only initialize once
|
328 |
+
if labels_data is None:
|
329 |
+
print("Downloading model files...") # Moved print here
|
330 |
model_path, tag_mapping_path = download_model_files()
|
331 |
+
model_path_global = model_path # Store the path
|
332 |
+
print("Loading labels...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
333 |
labels_data, _, tag_to_category_map = load_tag_mapping(tag_mapping_path)
|
334 |
+
print("Labels loaded.")
|
335 |
+
# --- Removed ONNX Session Initialization ---
|
336 |
|
337 |
@spaces.GPU()
|
338 |
def predict(image_input, gen_threshold, char_threshold, output_mode):
|
339 |
+
print("--- predict function started (GPU worker) ---")
|
340 |
+
"""Gradioインターフェース用の予測関数 (GPUワーカー内)"""
|
341 |
+
initialize_model() # Ensure files/labels are ready
|
342 |
+
|
343 |
+
# --- Create ONNX session inside the GPU function ---
|
344 |
+
print("Creating ONNX session for prediction...")
|
345 |
+
global model_path_global # Access the global model path
|
346 |
+
if model_path_global is None:
|
347 |
+
# Attempt initialization again if model path is missing (e.g., after restart)
|
348 |
+
initialize_model()
|
349 |
+
if model_path_global is None:
|
350 |
+
return "Error: Model path could not be initialized.", None
|
351 |
+
|
352 |
+
available_providers = ort.get_available_providers()
|
353 |
+
print(f"(Worker) Available ONNX Runtime providers: {available_providers}")
|
354 |
+
providers = []
|
355 |
+
if 'CUDAExecutionProvider' in available_providers:
|
356 |
+
providers.append('CUDAExecutionProvider')
|
357 |
+
providers.append('CPUExecutionProvider') # Always include CPU as fallback
|
358 |
+
|
359 |
+
try:
|
360 |
+
# Create session with GPU preference inside the worker
|
361 |
+
session = ort.InferenceSession(model_path_global, providers=providers)
|
362 |
+
print(f"(Worker) Using ONNX Runtime provider: {session.get_providers()[0]}")
|
363 |
+
except Exception as e:
|
364 |
+
print(f"(Worker) Error initializing ONNX session with providers {providers}: {e}")
|
365 |
+
# Fallback explicitly to CPU if GPU fails inside worker
|
366 |
+
try:
|
367 |
+
print("(Worker) Falling back to CPUExecutionProvider only.")
|
368 |
+
session = ort.InferenceSession(model_path_global, providers=['CPUExecutionProvider'])
|
369 |
+
except Exception as e_cpu:
|
370 |
+
print(f"(Worker) Error initializing ONNX session even with CPU: {e_cpu}")
|
371 |
+
return f"Error initializing ONNX session: {e_cpu}", None
|
372 |
+
# --- Session created ---
|
373 |
|
374 |
if image_input is None:
|
375 |
return "Please upload an image.", None
|
376 |
|
377 |
+
print(f"(Worker) Processing image with thresholds: gen={gen_threshold}, char={char_threshold}")
|
378 |
|
379 |
# PIL Imageオブジェクトであることを確認
|
380 |
if not isinstance(image_input, Image.Image):
|
381 |
+
try:
|
382 |
+
# URLの場合
|
383 |
+
if isinstance(image_input, str) and image_input.startswith("http"):
|
384 |
+
response = requests.get(image_input)
|
385 |
+
response.raise_for_status()
|
386 |
+
image = Image.open(io.BytesIO(response.content))
|
387 |
+
# ファイルパスの場合 (Gradioでは通常発生しないが念のため)
|
388 |
+
elif isinstance(image_input, str) and os.path.exists(image_input):
|
389 |
+
image = Image.open(image_input)
|
390 |
+
# Numpy配列の場合 (Gradio Imageコンポーネントからの入力)
|
391 |
+
elif isinstance(image_input, np.ndarray):
|
392 |
+
image = Image.fromarray(image_input)
|
393 |
+
else:
|
394 |
+
raise ValueError("Unsupported image input type")
|
395 |
+
except Exception as e:
|
396 |
+
print(f"(Worker) Error loading image: {e}")
|
397 |
+
return f"Error loading image: {e}", None
|
398 |
else:
|
399 |
image = image_input
|
400 |
|
|
|
401 |
# 前処理
|
402 |
original_pil_image, input_data = preprocess_image(image)
|
403 |
|
404 |
# データ型をモデルの期待に合わせる (通常はfloat32)
|
405 |
+
input_name = session.get_inputs()[0].name
|
406 |
+
expected_type = session.get_inputs()[0].type
|
407 |
if expected_type == 'tensor(float16)':
|
408 |
input_data = input_data.astype(np.float16)
|
409 |
else:
|
410 |
input_data = input_data.astype(np.float32) # Default to float32
|
411 |
|
412 |
+
# 推論 (作成したセッションを使用)
|
413 |
start_time = time.time()
|
414 |
+
outputs = session.run(None, {input_name: input_data})[0]
|
415 |
inference_time = time.time() - start_time
|
416 |
+
print(f"(Worker) Inference completed in {inference_time:.3f} seconds")
|
417 |
|
418 |
# シグモイド関数で確率に変換
|
419 |
probs = 1 / (1 + np.exp(-outputs[0])) # Apply sigmoid to the first batch item
|
|
|
427 |
if predictions["rating"]:
|
428 |
output_tags.append(predictions["rating"][0][0].replace("_", " "))
|
429 |
if predictions["quality"]:
|
430 |
+
output_tags.append(predictions["quality"][0][0].replace("_", " "))
|
431 |
|
432 |
# 残りのカテゴリをアルファベット順に追加(オプション)
|
433 |
for category in ["artist", "character", "copyright", "general", "meta"]:
|
434 |
tags = [tag.replace("_", " ") for tag, prob in predictions[category]
|
435 |
+
if not (category == "meta" and any(p in tag.lower() for p in ['id', 'commentary','mismatch']))] # メタタグフィルタリング
|
436 |
output_tags.extend(tags)
|
437 |
|
438 |
output_text = ", ".join(output_tags)
|
|
|
444 |
return output_text, viz_image
|
445 |
|
446 |
# --- Gradio Interface Definition ---
|
|
|
447 |
|
448 |
# CSS for styling
|
449 |
css = """
|
|
|
583 |
# 環境変数HF_TOKENがない場合に警告(プライベートリポジトリ用)
|
584 |
if not os.environ.get("HF_TOKEN"):
|
585 |
print("Warning: HF_TOKEN environment variable not set. Downloads from private repositories may fail.")
|
586 |
+
# initialize_model() # Removed startup initialization (model loaded in predict)
|
|
|
587 |
demo.launch(share=True)
|