Team: ECNU

Phase 1

Step 1. Follow the official instruction to install environment

Step 2. Dataset preparation

Make sure the nuScenes dataset can be accessed at: ./data/nuscenes

Step 3. Convert format

python convert_format_one_image.py robosense_track1_release.json robosense_track1_release_convert_one_image.json

Step 4. Start VLM service

bash service_kimi_vl.sh

Step 5. Inference

bash inference_one_image_kimi.sh

Step 6. Post-processing

# First, change the input_path to your inference result json in remove_think_tag.py*
# Then, execute:
python remove_think_tag.py 

Phase 2

Step 1. Follow the official instruction to install environment

Step 2. Dataset preparation

Make sure the nuScenes dataset can be accessed at: ./data/nuscenes, and the phase 2 dataset can be accessed at: ./data/robosense/phase2/samples

Step 3. Convert format

python convert_format_one_image_phase2.py robosense_track1_phase2.json robosense_track1_phase2_convert_one_image.json

Step 4. Start VLM service

bash service_kimi_vl.sh

Step 5. Inference

bash inference_one_image_kimi_phase2.sh

Step 6. Post-processing

# First, change the input_path to your inference result json in remove_think_tag.py*
# Then, execute:
python remove_think_tag.py