Fix usage instructions: correct model class / plugin guidance, English-only

README.md

@@ -1,32 +1,41 @@
 ---
 base_model: Qwen/Qwen3-VL-8B-Instruct
 license: apache-2.0
+library_name: transformers
+pipeline_tag: video-text-to-text
 tags:
 - camera-movement
 - video-understanding
 - qwen3-vl
-- sft
 - distillation
-library_name: transformers
-pipeline_tag: video-text-to-text
 ---
 
 # CamDistill-8B
 
-Camera-movement distillation SFT model, based on `Qwen/Qwen3-VL-8B-Instruct`.
+Camera-movement understanding model trained with **Camera Token Distillation** on top of
+`Qwen/Qwen3-VL-8B-Instruct`. A lightweight Camera Token Module learns geometry-aware camera
+tokens (distilled from VGGT) and injects them into the language model. Given a video, it outputs
+structured JSON describing every camera-movement segment.
 
-- Checkpoint: v5-20260724-054032 / checkpoint-1326
-- Training framework: [ms-swift](https://github.com/modelscope/ms-swift)
-- CamDistill config: DEPTH=8, EXTRACT_LAYERS=1,3,5,7,9,11,13,15, LAMBDA=0.05
+- **Paper**: [Temporally Grounded Compositional Camera Motion Understanding via Geometric Knowledge Distillation](https://huggingface.co/papers/2608.10932)
+- **Project page**: https://ddz16.github.io/cammotion.github.io
+- **Code**: https://github.com/ddz16/CamDistill
 
-This model is described in the paper [Temporally Grounded Compositional Camera Motion Understanding via Geometric Knowledge Distillation](https://huggingface.co/papers/2608.10932).  
-Project page: https://ddz16.github.io/cammotion.github.io  
-Code: https://github.com/ddz16/CamDistill
+> ⚠️ **This model cannot be loaded with plain 🤗 Transformers.** It contains an extra Camera Token
+> Module and a patched forward pass. Loading it as a standard `Qwen3VLForConditionalGeneration`
+> would silently drop those weights and produce incorrect results. Use the CamDistill repo, which
+> registers the required custom model type through a plugin.
 
 ## Usage
 
-```python
-from transformers import AutoModelForCausalLM, AutoProcessor
-model = AutoModelForCausalLM.from_pretrained("ddz16/CamDistill-8B", torch_dtype="bfloat16", device_map="auto")
-processor = AutoProcessor.from_pretrained("ddz16/CamDistill-8B")
-```
+Clone the [CamDistill repo](https://github.com/ddz16/CamDistill), then run (camera tokens are generated internally — **no online
+VGGT required**):
+
+```bash
+python camera_movement_sft/infer_single.py \
+    --model ddz16/CamDistill-8B \
+    --video /path/to/video.mp4 \
+    --variant camdistill
+```
+
+See the repo's README for environment setup and batch evaluation.