linjieli222/HERO

Datapath

avinashsai opened this issue · 5 comments

Hi,
Congrats on the amazing work.

I downloaded data using bash scripts/download_tvr.sh $PATH_TO_STORAGE (I gave my own $PATH_TO_STORAGE). How can I mention this data path in the fine-tuning script both single gpu and 8 gpus?

Thank you.

Thanks for your interests in this project. Please take a look at https://github.com/linjieli222/HERO#quick-start. When launching docker container, you will need to point to $PATH_TO_STORAGE

Hi,
Thanks for your reply. I want to run outside the docker. Can you point to the path change for it?

Take look at this file

--mount src=$OUTPUT,dst=/storage,type=bind \

Basically, we map
(1) $PATH_TO_STORAGE/txt_db to /txt inside the docker
(2) $PATH_TO_STORAGE/video_db to /video inside the docker
(3) $PATH_TO_STORAGE/finetune to /storage inside the docker
(4) $PATH_TO_STORAGE/pretrained to /pretrain inside the docker

You can change the finetuning command accordingly.

Thanks for your response. I am using single GPU instead of 8 GPUs. What are the changes to be made in the config file eg: train-tvc-8gpu.json to make sure that I am training for the same number of epochs, training samples on a single GPU?

The simple answer is 8 x the default value in config.

However, it is highly possible that your single-GPU finetuning cannot match our 8-GPU finetuning result. Check this thread on why: #14 (comment)