Polish the model card: badges, hero figure, and a tighter structure

README.md

@@ -12,28 +12,29 @@ library_name: pytorch
 pipeline_tag: video-classification
 ---
 
-# SurgRec — Surgical Video Pretraining Checkpoints
+# SurgRec — Surgical Video Foundation Models
 
-Pretrained backbones for
-[**Scaling Video Pretraining for Surgical Foundation Models**](https://arxiv.org/abs/2603.29966)
-(arXiv:2603.29966). Code, splits, and the launch scripts used for the reported
-runs live in
-[Siiichenggg/SurgRec-open-source](https://github.com/Siiichenggg/SurgRec-open-source).
+[![arXiv](https://img.shields.io/badge/arXiv-2603.29966-b31b1b.svg)](https://arxiv.org/abs/2603.29966)
+[![GitHub](https://img.shields.io/badge/GitHub-SurgRec--open--source-181717.svg?logo=github)](https://github.com/Siiichenggg/SurgRec-open-source)
+[![License: CC BY-NC 4.0](https://img.shields.io/badge/License-CC%20BY--NC%204.0-lightgrey.svg)](https://creativecommons.org/licenses/by-nc/4.0/)
 
-**Authors.** Sicheng Lu, Zikai Xiao, Jianhui Wei, Danyu Sun, Qi Lu, Keli Hu,
-Yang Feng, Jian Wu, Zongxin Yang, Zuozhu Liu.
+Pretrained backbones from
+[**Scaling Video Pretraining for Surgical Foundation Models**](https://arxiv.org/abs/2603.29966).
 
-These checkpoints are the product of that joint work. This repository is hosted
-under an individual account for distribution only; it does not indicate sole
-authorship.
+![Overview of the SurgRec pretraining and evaluation framework](overview.jpg)
 
-These are **pretrained backbones**, not task models. They carry no
-classification head and are meant to be fine-tuned on a downstream surgical
-video dataset.
+<p align="center"><em>Overview of the proposed surgical video pretraining and evaluation framework.</em></p>
 
-![Overview of the SurgRec pretraining and evaluation framework](overview.jpg)
+```bash
+hf download SichengLu/SurgRec --local-dir checkpoints
+```
+
+> These are **headless backbones** — no classification head. Fine-tune them on a
+> downstream surgical video dataset with the
+> [code repository](https://github.com/Siiichenggg/SurgRec-open-source).
 
-*Overview of the proposed surgical video pretraining and evaluation framework.*
+**Sicheng Lu, Zikai Xiao, Jianhui Wei, Danyu Sun, Qi Lu, Keli Hu, Yang Feng,
+Jian Wu, Zongxin Yang, Zuozhu Liu**
 
 ## Highlights
 
@@ -71,43 +72,41 @@ Top-1 accuracy (%) on the 16-dataset downstream benchmark, from Table 2 of the
 | SAR-RARP50 | 23.01 | 26.88 | 25.97 | 24.37 | 22.10 | **37.59** | 36.67 |
 | **Mean** | 32.56 | 33.87 | 28.25 | 35.92 | 34.02 | **45.28** | 39.11 |
 
-## Files
+## Models
 
-| File | Model | Architecture | Params | Objective |
-| --- | --- | --- | --- | --- |
-| `surgrec_mae.pth` | SurgRec-MAE | ViT-B/16, 12 blocks, width 768 | 94.2 M | Masked video reconstruction (VideoMAE-style) |
-| `surgrec_jepa.pt` | SurgRec-JEPA | ViT-L/16, 24 blocks, width 1024 | 303.9 M | Latent predictive pretraining (V-JEPA-style) |
-
-SHA256, for verifying downloads:
-
-```text
-ce2c0298292d636f95c77a12c26fcdeb93de8b4a78e515a68d247de5e61759c5  surgrec_mae.pth
-84040aa2e2fc015d03165903e6eec06282507f1d275e76efa97292457cd1bd3e  surgrec_jepa.pt
-```
+| File | Model | Architecture | Params | Pretraining objective | Load with |
+| --- | --- | --- | --- | --- | --- |
+| `surgrec_mae.pth` | SurgRec-MAE | ViT-B/16 | 94.2 M | Masked video reconstruction (VideoMAE-style) | `model_key=model` |
+| `surgrec_jepa.pt` | SurgRec-JEPA | ViT-L/16 | 303.9 M | Latent predictive (V-JEPA-style) | `model_key=encoder` |
 
-### `surgrec_mae.pth`
+<details>
+<summary><b>Checkpoint internals, training configuration, and checksums</b></summary>
 
-Recorded training configuration, read from the checkpoint itself:
+**`surgrec_mae.pth`** — configuration recorded inside the checkpoint:
 
 - `model = pretrain_videomae_base_patch16_224`
 - `mask_type = tube`, `mask_ratio = 0.9`, `decoder_depth = 4`
 - `input_size = 224`, `num_frames = 16`, `sampling_rate = 4`
-- `lr = 3e-4`, `batch_size = 64`
-- checkpoint taken at `epoch = 149`
+- `lr = 3e-4`, `batch_size = 64`, taken at `epoch = 149`
 
-The archive contains `model`, `optimizer`, `scaler`, `epoch`, and `args`. The
-`model` entry holds both the encoder and the MAE decoder; fine-tuning loads the
-encoder and discards `decoder.*`. Load it with `model_key = model`.
+The archive holds `model`, `optimizer`, `scaler`, `epoch`, and `args`. The
+`model` entry contains both the encoder and the MAE decoder; fine-tuning loads
+the encoder and discards `decoder.*`.
 
-### `surgrec_jepa.pt`
+**`surgrec_jepa.pt`** — holds `encoder`, `predictor`, and `target_encoder`, each
+ViT-L/16; taken at `epoch = 131`, `loss = 0.4225`, trained with
+`batch_size = 24` across `world_size = 8`. Load the `encoder` entry; its weights
+sit under a `module.backbone.` prefix, which the loader in the code repository
+strips.
 
-- `encoder`, `predictor`, and `target_encoder`, each ViT-L/16
-- checkpoint taken at `epoch = 131`, `loss = 0.4225`
-- trained with `batch_size = 24` across `world_size = 8`
+SHA256, for verifying downloads:
 
-Load the `encoder` entry for fine-tuning (`model_key = encoder`); weights are
-stored under a `module.backbone.` prefix, which the loader in the code
-repository strips.
+```text
+ce2c0298292d636f95c77a12c26fcdeb93de8b4a78e515a68d247de5e61759c5  surgrec_mae.pth
+84040aa2e2fc015d03165903e6eec06282507f1d275e76efa97292457cd1bd3e  surgrec_jepa.pt
+```
+
+</details>
 
 ## Usage
 
@@ -115,7 +114,6 @@ repository strips.
 git clone https://github.com/Siiichenggg/SurgRec-open-source.git
 cd SurgRec-open-source
 pip install -r requirements.txt
-
 hf download SichengLu/SurgRec --local-dir checkpoints
 ```
 
@@ -133,21 +131,15 @@ CKPT_ROOT=checkpoints DATA_ROOT=/path/to/split GPUS=8 \
   bash scripts/paper/finetune/jepa_v3_finetune_8gpu.sh
 ```
 
-## Provenance
-
-`surgrec_mae.pth` is released from `videomae_e149.pth` and `surgrec_jepa.pt`
-from `jepa_v3_60w_100e_20w_130e.pt`; the names follow `MODEL_ZOO.md` in the code
-repository. See the paper for the pretraining corpus and its sampling
-procedure.
-
-## License
+## License and intended use
 
-[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) — non-commercial
-use with attribution. The training code derives in part from VideoMAE, which
-carries the same obligation. See `LICENSE` and `NOTICE` in the code repository.
+Released for **non-commercial research**. These weights are a research artifact,
+not a medical device, and must not be used for clinical diagnosis or treatment
+decisions.
 
-These weights are a research artifact, not a medical device, and must not be
-used for clinical decisions.
+[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) — attribution
+required, no commercial use. The code derives in part from VideoMAE, which
+carries the same obligation; see `LICENSE` and `NOTICE` in the code repository.
 
 ## Citation