Upload folder using huggingface_hub

README.md

@@ -1,13 +1,13 @@
 ---
 title: SAM3 AutoTag
-emoji: 📉
-colorFrom: red
+emoji: 🏷️
+colorFrom: green
 colorTo: blue
 sdk: gradio
-sdk_version: 6.15.2
-python_version: '3.12'
+sdk_version: 6.6.0
 app_file: app.py
 pinned: false
+short_description: Florence-2 tagging for SAM 3 auto-discovery
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+`api_autotag(image, max_tags)` -> JSON {labels: [...], tags: [{name, count}]} via Florence-2 <OD>.

app.py

@@ -0,0 +1,45 @@
+"""Florence-2 object tagger (ZeroGPU): proposes class names from an image."""
+import os
+from collections import Counter
+
+import gradio as gr
+import spaces
+import torch
+from transformers import AutoProcessor, Florence2ForConditionalGeneration
+
+MODEL_ID = "florence-community/Florence-2-large"
+
+# Built at import on CPU; moved to CUDA inside the @spaces.GPU function.
+processor = AutoProcessor.from_pretrained(MODEL_ID)
+model = Florence2ForConditionalGeneration.from_pretrained(MODEL_ID)
+model.eval()
+
+
+@spaces.GPU(duration=120)
+def api_autotag(image, max_tags):
+    if image is None:
+        return {"error": "no image provided"}
+    image = image.convert("RGB")
+    device = "cuda"
+    model.to(device)
+    inputs = processor(text="<OD>", images=image, return_tensors="pt").to(device)
+    with torch.no_grad():
+        gen = model.generate(**inputs, max_new_tokens=1024, num_beams=3)
+    text = processor.batch_decode(gen, skip_special_tokens=False)[0]
+    parsed = processor.post_process_generation(text, task="<OD>", image_size=image.size)
+    labels = parsed.get("<OD>", {}).get("labels", [])
+    counts = Counter(str(l).strip().lower() for l in labels if str(l).strip())
+    tags = [{"name": n, "count": c} for n, c in counts.most_common(int(max_tags))]
+    return {"model": MODEL_ID, "tags": tags, "labels": [t["name"] for t in tags]}
+
+
+with gr.Blocks(title="SAM3 AutoTag") as demo:
+    gr.Markdown("# Florence-2 AutoTag\nUpload an image; returns detected object class names.")
+    with gr.Row():
+        inp = gr.Image(type="pil", label="Image")
+        out = gr.JSON(label="Tags")
+    mt = gr.Slider(1, 50, value=20, step=1, label="Max tags")
+    gr.Button("Tag").click(api_autotag, [inp, mt], out, api_name="api_autotag")
+
+if __name__ == "__main__":
+    demo.queue().launch(show_error=True)

requirements.txt

@@ -0,0 +1,9 @@
+transformers==5.9.0
+torch==2.11.0
+torchvision
+gradio==6.6.0
+spaces
+timm
+einops
+pillow
+numpy