rese1f / MovieChat

[CVPR 2024] MovieChat: From Dense Token to Sparse Memory for Long Video Understanding
https://rese1f.github.io/MovieChat/
BSD 3-Clause "New" or "Revised" License
534 stars 41 forks source link

YAML file #20

Closed fazliimam closed 1 year ago

fazliimam commented 1 year ago

In the yaml file for llama_model should it be the --target directory of the apply_delta.py or the --delta

Espere-1119-Song commented 1 year ago

It should be the --target directory :)

fazliimam commented 1 year ago

Noted, thank you. On another note what is the --fragment-video-path argument for inference.py?

Espere-1119-Song commented 1 year ago

Since the video is relatively long and cannot be read all at once, we set up a sliding window to read the video step by step. --fragment-video-path stores the video read by the sliding window at each time step.