Add files

app.py

@@ -0,0 +1,161 @@
+#!/usr/bin/env python
+
+from __future__ import annotations
+
+import argparse
+import functools
+import json
+import os
+import pathlib
+import subprocess
+from typing import Callable
+
+# workaround for https://github.com/gradio-app/gradio/issues/483
+command = 'pip install -U gradio==2.7.0'
+subprocess.call(command.split())
+
+import gradio as gr
+import huggingface_hub
+import PIL.Image
+import torch
+import torchvision.transforms as T
+
+TOKEN = os.environ['TOKEN']
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--device', type=str, default='cpu')
+    parser.add_argument('--score-slider-step', type=float, default=0.05)
+    parser.add_argument('--score-threshold', type=float, default=0.4)
+    parser.add_argument('--theme', type=str, default='dark-grass')
+    parser.add_argument('--live', action='store_true')
+    parser.add_argument('--share', action='store_true')
+    parser.add_argument('--port', type=int)
+    parser.add_argument('--disable-queue',
+                        dest='enable_queue',
+                        action='store_false')
+    parser.add_argument('--allow-flagging', type=str, default='never')
+    parser.add_argument('--allow-screenshot', action='store_true')
+    return parser.parse_args()
+
+
+def download_sample_images() -> list[pathlib.Path]:
+    image_dir = pathlib.Path('samples')
+    image_dir.mkdir(exist_ok=True)
+
+    dataset_repo = 'hysts/sample-images-TADNE'
+    n_images = 36
+    paths = []
+    for index in range(n_images):
+        path = huggingface_hub.hf_hub_download(dataset_repo,
+                                               f'{index:02d}.jpg',
+                                               repo_type='dataset',
+                                               cache_dir=image_dir.as_posix(),
+                                               use_auth_token=TOKEN)
+        paths.append(pathlib.Path(path))
+    return paths
+
+
+@torch.inference_mode()
+def predict(image: PIL.Image.Image, score_threshold: float,
+            transform: Callable, device: torch.device, model: torch.nn.Module,
+            labels: list[str]) -> dict[str, float]:
+    data = transform(image)
+    data = data.to(device).unsqueeze(0)
+    preds = model(data)[0]
+    preds = torch.sigmoid(preds)
+    preds = preds.cpu().numpy().astype(float)
+
+    res = dict()
+    for prob, label in zip(preds, labels):
+        if prob < score_threshold:
+            continue
+        res[label] = prob
+    return res
+
+
+def load_labels() -> list[str]:
+    label_path = pathlib.Path('class_names_6000.json')
+    label_url = 'https://raw.githubusercontent.com/RF5/danbooru-pretrained/master/config/class_names_6000.json'
+    if not label_path.exists():
+        torch.hub.download_url_to_file(label_url, label_path.as_posix())
+    with open(label_path) as f:
+        labels = json.load(f)
+    return labels
+
+
+def main():
+    gr.close_all()
+
+    args = parse_args()
+    device = torch.device(args.device)
+
+    image_paths = download_sample_images()
+    examples = [[path.as_posix(), args.score_threshold]
+                for path in image_paths]
+
+    if device.type == 'cpu':
+        model_path = pathlib.Path('resnet50-13306192.pth')
+        model_url = 'https://github.com/RF5/danbooru-pretrained/releases/download/v0.1/resnet50-13306192.pth'
+        if not model_path.exists():
+            torch.hub.download_url_to_file(model_url, model_path.as_posix())
+        model = torch.hub.load('RF5/danbooru-pretrained',
+                               'resnet50',
+                               pretrained=False)
+        state_dict = torch.load(model_path, map_location=device)
+        model.load_state_dict(state_dict)
+    else:
+        model = torch.hub.load('RF5/danbooru-pretrained', 'resnet50')
+    model.to(device)
+    model.eval()
+
+    transform = T.Compose([
+        T.Resize(360),
+        T.ToTensor(),
+        T.Normalize(mean=[0.7137, 0.6628, 0.6519],
+                    std=[0.2970, 0.3017, 0.2979]),
+    ])
+
+    labels = load_labels()
+
+    func = functools.partial(predict,
+                             transform=transform,
+                             device=device,
+                             model=model,
+                             labels=labels)
+    func = functools.update_wrapper(func, predict)
+
+    repo_url = 'https://github.com/RF5/danbooru-pretrained'
+    title = 'RF5/danbooru-pretrained'
+    description = f'A demo for {repo_url}'
+    article = None
+
+    gr.Interface(
+        func,
+        [
+            gr.inputs.Image(type='pil', label='Input'),
+            gr.inputs.Slider(0,
+                             1,
+                             step=args.score_slider_step,
+                             default=args.score_threshold,
+                             label='Score Threshold'),
+        ],
+        gr.outputs.Label(label='Output'),
+        theme=args.theme,
+        title=title,
+        description=description,
+        article=article,
+        examples=examples,
+        allow_screenshot=args.allow_screenshot,
+        allow_flagging=args.allow_flagging,
+        live=args.live,
+    ).launch(
+        enable_queue=args.enable_queue,
+        server_port=args.port,
+        share=args.share,
+    )
+
+
+if __name__ == '__main__':
+    main()

requirements.txt

@@ -0,0 +1,2 @@
+torch>=1.10.1
+torchvision>=0.11.2