---
license: apache-2.0
---

## 🔥 Reproduce Website Demos

1. **[Environment Set Up]** Our environment setup is identical to [CogVideoX](https://github.com/THUDM/CogVideo). You can refer to their configuration to complete the environment setup.
    ```bash
    conda create -n robomaster python=3.10
    conda activate robomaster
    ```

2. Robotic Manipulation on Diverse Out-of-Domain Objects.
    ```bash
    python inference_inthewild.py \
        --input_path demos/diverse_ood_objs \
        --output_path samples/infer_diverse_ood_objs \
        --transformer_path ckpts/RoboMaster \
        --model_path ckpts/CogVideoX-Fun-V1.5-5b-InP
    ```

3. Robotic Manipulation with Diverse Skills
    ```bash
    python inference_inthewild.py \
        --input_path demos/diverse_skills \
        --output_path samples/infer_diverse_skills \
        --transformer_path ckpts/RoboMaster \
        --model_path ckpts/CogVideoX-Fun-V1.5-5b-InP
    ```

4. Long Video Generation in Auto-Regressive Manner
    ```bash
    python inference_inthewild.py \
        --input_path demos/long_video \
        --output_path samples/long_video \
        --transformer_path ckpts/RoboMaster \
        --model_path ckpts/CogVideoX-Fun-V1.5-5b-InP
    ```
    
## 🚀 Benchmark Evaluation (Reproduce Paper Results)
  ```
├── RoboMaster
    ├── eval_metrics
        ├── VBench
        ├── common_metrics_on_video_quality
        ├── eval_traj
        ├── results
            ├── bridge_eval_gt
            ├── bridge_eval_ours
            ├── bridge_eval_ours_tracking
  ```

**(1) Inference on Benchmark & Prepare Evaluation Files**
1. Generating `bridge_eval_ours`. (Note that the results may vary slightly across different computing machines, even with the same seed. We have prepared the reference files under `eval_metrics/results`)
    ```bash
    cd RoboMaster/
    python inference_eval.py
    ```
1. Generating `bridge_eval_ours_tracking`: Install [CoTracker3](https://github.com/facebookresearch/co-tracker), and then estimate tracking points with grid size 30 on `bridge_eval_ours`. 

**(2) Evaluation on Visual Quality**

1. Evaluation of VBench metrics.
    ```bash
    cd eval_metrics/VBench
    python evaluate.py \
        --dimension aesthetic_quality imaging_quality temporal_flickering motion_smoothness subject_consistency background_consistency \
        --videos_path ../results/bridge_eval_ours \
        --mode=custom_input \
        --output_path evaluation_results
    ```

2. Evaluation of FVD and FID metrics.
    ```bash
    cd eval_metrics/common_metrics_on_video_quality
    python calculate.py -v1_f ../results/bridge_eval_ours -v2_f ../results/bridge_eval_gt
    python -m pytorch_fid eval_1 eval_2
    ```


**(3) Evaluation on Trajectory (Robotic Arm & Manipulated Object)**

1. Estimation of TrajError metrics. (Note that we exclude some samples listed in `failed_track.txt`, due to failed estimation by [CoTracker3](https://github.com/facebookresearch/co-tracker))
    ```bash
    cd eval_metrics/eval_traj
    python calculate_traj.py \
        --input_path_1 ../results/bridge_eval_ours \
        --input_path_2 ../results/bridge_eval_gt \
        --tracking_path ../results/bridge_eval_ours_tracking \
        --output_path evaluation_results
    ```

2. Check the visualization videos under `evaluation_results`. We blend the trajectories of robotic arm and object throughout the entire video for better illustration.