search

minghangz / TFVTG

13 stars 1 forks source link

How to test on a custom dataset #1

Open Shi-Yuhong opened 2 months ago

Shi-Yuhong commented 2 months ago

Thank you for your excellent work. I have some questions that I hope to receive your answers to. I hope to apply TFVTG to my custom video dataset to test the video temporal grounding function. What should I need to do?

Shi-Yuhong commented 2 months ago

Thank you for your excellent work. I have some questions that I hope to receive your answers to. I hope to apply TFVTG to my custom video dataset to test the video temporal grounding function. What should I need to do?

Especially how to automatically obtain _llmoutputs.json. The code provided by the project does not seem to include the generation of this file.

minghangz commented 2 months ago

Thank you for being interested in our work. I will clean up and release the code for generating llm_outputs.json soon. In the meantime, you can use feature_extraction.py to extract video features from your custom dataset and test directly with the VLM localizer (run evaluate.py without specifying --llm_output). You may need to adjust the stride and max_stride_factor in data_configs.py to achieve better performance.

Shi-Yuhong commented 2 months ago

Thank you very much for your response. It has been extremely helpful for our work.

minghangz commented 2 months ago

Hi, I have uploaded get_llm_outputs.py and updated the README.