search

dvlab-research / LLaMA-VID

Official Implementation for LLaMA-VID: An Image is Worth 2 Tokens in Large Language Models
Apache License 2.0
623 stars 40 forks source link

Failed to run finetuning stage2 with ActivityNet videos #23

Closed XenonLamb closed 6 months ago

XenonLamb commented 6 months ago

Great work! I have an issue when trying to run the short video finetuning stage 2: Although I manually checked the ActivityNet videos I downloaded are in mp4 format and is playable, I received the errors like Error in loading 374062, retrying... Error in loading 370984, retrying... Error in loading 416357, retrying... Error in loading 407613, retrying... Error in loading 397054, retrying... Error in loading 400911, retrying... [h264 @ 0x561a1edabac0] Missing reference picture, default is 65530 [h264 @ 0x561a4b9d0f40] mmco: unref short failure [h264 @ 0x561a4b9d0f40] mmco: unref short failure [h264 @ 0x561a1edabac0] Missing reference picture, default is 65530 [h264 @ 0x561a4b9d0f40] mmco: unref short failure [h264 @ 0x561a4b9d0f40] mmco: unref short failure [h264 @ 0x55c33946bfc0] mmco: unref short failure [h264 @ 0x55c33946bfc0] mmco: unref short failure [h264 @ 0x55c33946bfc0] mmco: unref short failure [h264 @ 0x55c33946bfc0] mmco: unref short failure Error in loading 414902, retrying... Error in loading 375344, retrying... I'm not very familiar with the video decoding protocols. Any suggestions on how this happens and what we should to to fix it is appreciated. Thank you!

yanwei-li commented 6 months ago

Hi, if there are too many warnings like Error in loading xxxx, it could be strange and you may need to debug to find if the video file is missing. And for the warning [h264 @ 0x561a4b9d0f40] mmco: unref short failure, we also find such warnings in video processing and we just ignore it.