Inference E.T. Chat on Charades-STA

The performance of E.T. Chat on full Charades-STA test set is as follows.

Model	R1@0.3	R1@0.5	R1@0.7	mIoU
E.T. Chat	65.7	45.9	20.0	42.3

📦 Data Preparation

Download Charades-STA dataset and organize it as follows.

ETBench
└─ data
   └─ charades
      ├─ videos
      │  ├─ 0A8CF.mp4
      │  ├─ 0A8ZT.mp4
      │  └─ ...
      ├─ charades_sta_train.txt
      └─ charades_sta_test.txt

🚀 Run Inference

Run the following command to inference E.T. Chat on Charades-STA dataset.

python etchat/eval/infer_charades.py \
    --anno_path data/charades/charades_sta_test.txt \
    --data_path data/charades/videos \
    --pred_path charades_sta \
    --model_path <path-to-model>

The model outputs would be saved in the charades_sta folder.

🔮 Compute Metrics

Run the following command to compute metrics.

python etchat/eval/eval_grounding.py charades_sta