Update README.md

README.md

@@ -10,138 +10,38 @@ tags:
 - video
 - video-summarization
 - highlight-detection
-- temporal-annotations
 homepage: https://ipcv.github.io/SoccerHigh/
 repository: https://github.com/IPCV/SoccerHigh
 ---
 
 # ⚽ SoccerHigh
 
-This code provides the implementation of the **Baseline Model** introduced in:
+This model card provides the checkpoints for the **Baseline Model** introduced in:
 
 [**SoccerHigh: A Benchmark Dataset for Automatic Soccer Video Summarization**](https://dl.acm.org/doi/pdf/10.1145/3728423.3759410)  
 Artur Díaz-Juan, Coloma Ballester, Gloria Haro  
 ACM MMSports 2025
 
----
-
-## 📂 Code Structure
+## 📂 Model Checkpoints
 
-The code for the baseline model is organized as follows:
-
-```text
-code/
-├── augmentations/           
-│   ├── mixup.py                          # Mixup augmentation implementation
-├── configs/                              # Hydra configuration files
-|   ├── datamodules/
-│   |   ├── default.yaml
-|   ├── datasets/
-│   |   ├── default.yaml
-|   ├── models/
-│   |   ├── default.yaml
-|   ├── scripts/
-│   |   ├── trim_summary.yaml
-|   ├── trainer/
-│   |   ├── default.yaml
-|   ├── predict.yaml
-|   ├── test.yaml
-│   ├── train.yaml
-├── datamodules/             
-│   ├── datamodule.py                     # Data module handling datasets and dataloaders
-├── datasets/                
-│   ├── soccernet_games.py                # Handles SoccerNet game-level data
-│   ├── soccernet_summarization.py        # Handles video summarization data
-│   ├── utils.py                          # Utility functions for dataset processing
-├── evaluation/              
-│   ├── evaluate.py                       # Compute metrics and summarize model performance
-├── inference/               
-│   ├── inference.py                      # Key shot selection for inference
-│   ├── utils.py                          # Helper functions for inference
-├── models/                  
-│   ├── classifier.py                     # Baseline model implementation
-│   ├── dino.py                           # DINO implementation
-│   ├── heads.py                          # Model heads definitions
-│   ├── transnetv2.py                     # TransNetv2 implementation
-├── scripts/                 
-│   ├── trimm_summary.py                  # Script to compute new summary annotations
-├── weights/
-│   ├── Baseline_VideoMAEv2-Giant.ckpt    # Baseline checkpoint using the VideoMAEv2 giant backbone 
-│   ├── Baseline_VideoMAEv2-Small.ckpt    # Baseline checkpoint using the VideoMAEv2 small backbone       
-├── predict.py                            # Inference script
-├── test.py                               # Testing script
-├── train.py                              # Training script
-```
+| Name | Description |
+|------|-------------|
+| Baseline_VideoMAEv2-Giant | Baseline model trained with a VideoMAEv2 encoder for feature extraction using [*vit_g_hybrid_pt_1200e_k710_ft*](https://huggingface.co/OpenGVLab/VideoMAE2/resolve/main/distill/vit_s_k710_dl_from_giant.pth) weights |
+| Baseline_VideoMAEv2-Small | Baseline model trained with a VideoMAEv2 encoder for feature extraction using [*vit_g_hybrid_pt_1200e_k710_ft*](https://github.com/OpenGVLab/VideoMAEv2/blob/master/docs/MODEL_ZOO.md) weights |
+| Baseline_CLIP | Baseline model trained with a CLIP encoder for feature extraction using [*CLIP ViT-B-32 256×256 trained on DataComp-1B*](https://huggingface.co/laion/CLIP-ViT-B-32-256x256-DataComp-s34B-b86K) weights |
+| Baseline_ResNet | Baseline model trained with a ResNet-152 encoder for feature extraction using [*IMAGENET1K_V2*](https://docs.pytorch.org/vision/main/models/generated/torchvision.models.resnet152.html#torchvision.models.ResNet152_Weights) weights |
 
 ---
 
-## ⚙️ Preparation
-
-Before running the code, set up the environment and the dataset path.
-
-**1. Create the Conda environment**:
-
-We provide an `environment.yml` file to install all dependencies:
-
-```bash
-conda env create -f environment.yml
-conda activate soccerhigh
-```
-
-**2. Create a symbolic link to the dataset**:
-
-The code requires a folder named `data` (either a symbolic link or the dataset itself), containing the files from [**SoccerHigh**](https://github.com/IPCV/SoccerHigh/tree/main/dataset).
+## ⚙️ Installation
 
 ```bash
-ln -s /path/to/your/dataset data
+git clone https://github.com/IPCV/SoccerHigh.git
+cd SoccerHigh/code
 ```
 
-Replace `/path/to/your/dataset` with the actual path to your dataset folder.
-
----
-
-## 🚀 Usage
-
-The code is designed using [PyTorch Lightning](https://lightning.ai/) for cleaner training loops and easier scaling.
-All configurations are managed using a [Hydra](https://hydra.cc/), making it easy to modify parameters.
-
-### 1. Training
-
-Train the model using the default configuration:
-
-```bash
-python3 train.py
-```
-
-### 2. Testing/Evaluation
-
-Evaluate a trained model checkpoint:
-
-```bash
-python3 test.py checkpoint='weights/checkpoint.ckpt'
-```
-
-Replace `checkpoint.ckpt` with the actual checkpoint name. Default: [`Baseline_VideoMAEv2-Giant.ckpt`](https://github.com/IPCV/SoccerHigh/blob/main/code/weights/Baseline_VideoMAEv2-Giant.ckpt)
-
-### 3. Inference
-
-Run inference on new games:
-
-```bash
-python3 predict.py checkpoint='weights/checkpoint.ckpt' output_path='file.json' datamodule.predict.dataset.game_list='/data/games.txt'
-```
-
-Replace `checkpoint.ckpt`, `file.json` and `games.txt` with the actual file names. Default: [`Baseline_VideoMAEv2-Giant.ckpt`](https://github.com/IPCV/SoccerHigh/blob/main/code/weights/Baseline_VideoMAEv2-Giant.ckpt), `output.json` and `test.txt`
-
----
-
-## 🔗 Notes
-
-- Make sure the symbolic link `data` points correctly to your dataset.
-
-- All configuration files are in `configs/` for easy experiment management.
-
-- Logs and checkpoints will be saved in the default generated directory `lightning_logs/`.
+For more details, please visit the original repository: 
+[https://github.com/IPCV/SoccerHigh](https://github.com/IPCV/SoccerHigh)
 
 ---