feat: update gradio build

Dockerfile

@@ -0,0 +1,23 @@
+FROM python:3.13.5-slim
+
+WORKDIR /app
+
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    curl \
+    git \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt ./
+RUN --mount=from=ghcr.io/astral-sh/uv:0.9.5,source=/uv,target=/bin/uv \
+    --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=src.txt,target=requirements.txt \
+    UV_TORCH_BACKEND='cpu' uv pip install --system -r requirements.txt
+
+COPY src/ ./src/
+
+EXPOSE 8501
+
+HEALTHCHECK CMD curl --fail http://localhost:8501
+
+ENTRYPOINT ["python", "src/gradio_app.py", "--port=3000", "--repo=frgfm/rexnet1_0x"]

README.md

@@ -3,10 +3,13 @@ title: Holocron
 emoji: 👁️‍🗨️
 colorFrom: gray
 colorTo: purple
-sdk: gradio
-app_file: app.py
-pinned: false
+sdk: docker
+app_port: 8501
+tags:
+  - gradio
+pinned: true
 license: apache-2.0
+short_description: PyTorch implementations of recent Computer Vision tricks (ReXNet, RepVGG, Unet3p, YOLOv4, CIoU loss, AdaBelief, PolyLoss, MobileOne).
 ---
 
 # Configuration

app.py

@@ -1,71 +0,0 @@
-# Copyright (C) 2022, François-Guillaume Fernandez.
-
-# This program is licensed under the Apache License 2.0.
-# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0> for full license details.
-
-import argparse
-import json
-
-import gradio as gr
-import numpy as np
-import onnxruntime
-from huggingface_hub import hf_hub_download
-from PIL import Image
-
-
-REPO = "frgfm/rexnet1_0x"
-
-# Download model config & checkpoint
-with open(hf_hub_download(REPO, filename="config.json"), "rb") as f:
-    cfg = json.load(f)
-
-ort_session = onnxruntime.InferenceSession(hf_hub_download(REPO, filename="model.onnx"))
-
-def preprocess_image(pil_img: Image.Image) -> np.ndarray:
-    """Preprocess an image for inference
-
-    Args:
-        pil_img: a valid pillow image
-
-    Returns:
-        the resized and normalized image of shape (1, C, H, W)
-    """
-
-    # Resizing (PIL takes (W, H) order for resizing)
-    img = pil_img.resize(cfg["input_shape"][-2:][::-1], Image.BILINEAR)
-    # (H, W, C) --> (C, H, W)
-    img = np.asarray(img).transpose((2, 0, 1)).astype(np.float32) / 255
-    # Normalization
-    img -= np.array(cfg["mean"])[:, None, None]
-    img /= np.array(cfg["std"])[:, None, None]
-
-    return img[None, ...]
-
-def predict(image):
-    # Preprocessing
-    np_img = preprocess_image(image)
-    ort_input = {ort_session.get_inputs()[0].name: np_img}
-
-    # Inference
-    ort_out = ort_session.run(None, ort_input)
-    # Post-processing
-    out_exp = np.exp(ort_out[0][0])
-    probs = out_exp / out_exp.sum()
-
-    return {class_name: float(conf) for class_name, conf in zip(cfg["classes"], probs)}
-
-img = gr.inputs.Image(type="pil")
-outputs = gr.outputs.Label(num_top_classes=3)
-
-gr.Interface(
-    fn=predict,
-    inputs=[img],
-    outputs=outputs,
-    title="Holocron: image classification demo",
-    article=(
-        "<p style='text-align: center'><a href='https://github.com/frgfm/Holocron'>"
-        "Github Repo</a> | "
-        "<a href='https://frgfm.github.io/Holocron/'>Documentation</a></p>"
-    ),
-    live=True,
-).launch()

requirements.txt

@@ -1,5 +1,5 @@
-gradio>=3.0.2,<4.0.0
-Pillow>=8.4.0
-onnxruntime>=1.10.0,<2.0.0
-huggingface-hub>=0.4.0,<1.0.0
-numpy>=1.19.5,<2.0.0
+gradio>=5.0.0,<6.0.0
+huggingface-hub>=1.0.0,<2.0.0
+numpy>=1.19.5,<3.0.0
+onnxruntime>=1.22.0,<2.0.0
+Pillow>=8.4.0,!=9.2.0

src/gradio_app.py

@@ -0,0 +1,91 @@
+# Copyright (C) 2022-2025, François-Guillaume Fernandez.
+
+# This program is licensed under the Apache License 2.0.
+# See LICENSE or go to <https://www.apache.org/licenses/LICENSE-2.0> for full license details.
+
+# /// script
+# requires-python = ">=3.11"
+# dependencies = [
+#     "gradio>=5.0.0,<6.0.0",
+#     "huggingface-hub>=1.0.0,<2.0.0",
+#     "numpy>=1.19.5,<3.0.0",
+#     "onnxruntime>=1.22.0,<2.0.0",
+#     "Pillow>=8.4.0,!=9.2.0",
+# ]
+# ///
+
+import json
+from argparse import ArgumentDefaultsHelpFormatter, ArgumentParser
+from pathlib import Path
+
+import gradio as gr
+import numpy as np
+import onnxruntime
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+
+def main(args):
+    # Download model config & checkpoint
+    with Path(hf_hub_download(args.repo, filename="config.json")).open("rb") as f:
+        cfg = json.load(f)
+
+    ort_session = onnxruntime.InferenceSession(hf_hub_download(args.repo, filename="model.onnx"))
+
+    def preprocess_image(pil_img: Image.Image) -> np.ndarray:
+        """Preprocess an image for inference
+
+        Args:
+            pil_img: a valid pillow image
+
+        Returns:
+            the resized and normalized image of shape (1, C, H, W)
+        """
+        # Resizing (PIL takes (W, H) order for resizing)
+        img = pil_img.resize(cfg["input_shape"][-2:][::-1], Image.BILINEAR)
+        # (H, W, C) --> (C, H, W)
+        img = np.asarray(img).transpose((2, 0, 1)).astype(np.float32) / 255
+        # Normalization
+        img -= np.array(cfg["mean"])[:, None, None]
+        img /= np.array(cfg["std"])[:, None, None]
+
+        return img[None, ...]
+
+    def predict(image):
+        # Preprocessing
+        np_img = preprocess_image(image)
+        ort_input = {ort_session.get_inputs()[0].name: np_img}
+
+        # Inference
+        ort_out = ort_session.run(None, ort_input)
+        # Post-processing
+        out_exp = np.exp(ort_out[0][0])
+        probs = out_exp / out_exp.sum()
+
+        return {class_name: float(conf) for class_name, conf in zip(cfg["classes"], probs, strict=True)}
+
+    interface = gr.Interface(
+        fn=predict,
+        inputs=gr.Image(type="pil"),
+        outputs=gr.Label(num_top_classes=3),
+        title="Holocron: image classification demo",
+        article=(
+            "<p style='text-align: center'><a href='https://github.com/frgfm/holocron'>"
+            "Github Repo</a> | "
+            "<a href='https://frgfm.github.io/holocron/'>Documentation</a></p>"
+        ),
+        live=True,
+    )
+
+    interface.launch(server_port=args.port, show_error=True)
+
+
+if __name__ == "__main__":
+    parser = ArgumentParser(
+        description="Holocron image classification demo", formatter_class=ArgumentDefaultsHelpFormatter
+    )
+    parser.add_argument("--repo", type=str, default="frgfm/rexnet1_0x", help="HF Hub repo to use")
+    parser.add_argument("--port", type=int, default=8001, help="Port on which the webserver will be run")
+    args = parser.parse_args()
+
+    main(args)