add vpp2 pretrained video model in stage1 and stage2 (#1)

- Add VPP2 Stage-1 49-frame video checkpoint (b5f0faf66df9fc53ab52a046d5414ff949ff5248)
- Add VPP2 Stage-2 17-frame video checkpoint (a9f66dd6bc79471e27d1eb66c7abaa6a4ab255f2)


Co-authored-by: Yanjiang Guo <yjguo@users.noreply.huggingface.co>

README.md

@@ -12,7 +12,7 @@ base_model: Wan-AI/Wan2.1-I2V-14B-480P
 
 # Video Prediction Policy 2: Predict Better, Act Better
 
-Official VPP2 weights for RoboDojo and LIBERO.
+Official VPP2 weights for zero-shot video prediction, RoboDojo and LIBERO.
 
 [Paper](https://arxiv.org/abs/2610.10270) · [Code](https://github.com/roboterax/video-prediction-policy-2) · [Project page](https://robert-gyj.github.io/video-prediction-policy-2/) · [ModelScope](https://modelscope.cn/models/haodong123/VPP2_preview)
 
@@ -20,6 +20,8 @@ Official VPP2 weights for RoboDojo and LIBERO.
 
 | Checkpoint | Path | Use |
 |---|---|---|
+| VPP2 Stage-1 Video (49 frames) | `checkpoints_video/vpp2-video-stage1-49f.pth` | Zero-shot image-to-video prediction |
+| VPP2 Stage-2 Video (17 frames) | `checkpoints_video/vpp2-video-stage2-17f.pth` | Zero-shot image-to-video prediction |
 | RoboDojo history-conditioned Video-10k | `checkpoints/initialization/robodojo_his10k.pt` | Joint + Action2B training initializer |
 | RoboDojo joint Video-100k | `checkpoints/joint2b_s100000/video.pt` | Paired RoboDojo evaluation |
 | RoboDojo Action2B-100k | `checkpoints/joint2b_s100000/action.pt` | Paired RoboDojo evaluation |
@@ -32,12 +34,34 @@ normalization statistics. Keep each pair together.
 
 ## TODO
 
-- [ ] Release the robot-video pretrained backbone checkpoint.
+- [ ] Merge the Stage-1 and Stage-2 video checkpoint release ([PR #1](https://huggingface.co/Haodong082399/VPP2/discussions/1)).
 
 ## Download
 
+Install the Hugging Face CLI:
+
 ```bash
 python -m pip install -U huggingface_hub
+```
+
+The Stage-1 and Stage-2 video checkpoints are available on PR #1 while the release awaits merging:
+
+```bash
+hf download Haodong082399/VPP2 \
+  checkpoints_video/vpp2-video-stage1-49f.pth \
+  checkpoints_video/vpp2-video-stage2-17f.pth \
+  --revision refs/pr/1 --local-dir weights
+```
+
+Use 49 frames for Stage-1 or 17 frames for Stage-2. Both checkpoints use regular
+denoising with 30 steps, CFG=4, and sigma shift 3 for standalone video prediction.
+The shared Wan encoders and tokenizer listed above are also required. Follow the
+[video prediction guide](https://github.com/roboterax/video-prediction-policy-2/blob/main/docs/video_prediction.md)
+for image preparation, prompt enhancement, and inference.
+
+To download the checkpoints on the main branch:
+
+```bash
 hf download Haodong082399/VPP2 --local-dir weights
 ```
 
@@ -48,7 +72,7 @@ hf download Haodong082399/VPP2 --local-dir weights \
   --include 'checkpoints/joint2b_s100000/*' --include 'checkpoints/Wan2.1-I2V-14B-480P/*'
 ```
 
-The directory structure matches the ModelScope release. Follow the official
+The RoboDojo and LIBERO directory structure matches the ModelScope release. Follow the official
 [RoboDojo guide](https://github.com/roboterax/video-prediction-policy-2/blob/main/docs/robodojo.md)
 or [LIBERO guide](https://github.com/roboterax/video-prediction-policy-2/blob/main/docs/libero.md)
 for installation, training and evaluation.

checkpoints_video/vpp2-video-stage1-49f.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c39effb224318c24584b07c866f465d1d6e080cf79fbc812aba5f3388e08f9f
+size 65580046353

checkpoints_video/vpp2-video-stage2-17f.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7c3efbbc8c3ebe71bbede9794410411fa624c6d5fc812a36f3496e53f780bf8
+size 65580046353