search

megvii-research / BBN

The official PyTorch implementation of paper BBN: Bilateral-Branch Network with Cumulative Learning for Long-Tailed Visual Recognition
https://arxiv.org/abs/1912.02413
MIT License
657 stars 100 forks source link

Training Time for iNaturalist'18 #8

Closed duggalrahul closed 4 years ago

duggalrahul commented 4 years ago

Hi,

Thanks for the nice work in this repo!

Could you share your GPU setup and the training time per epoch on iNaturalist'18?

Best, Rahul

ZhouBoyan commented 4 years ago

We use four NVIDIA 1080ti GPU cards in iNaturalist experiments and the CUDA nad CUDNN version is 9.0 and 7.1.3 respectively. With two batches of data sampled from dual samplers, our training time is more than the baseline experiments (26min v.s 15min). Some tricks could accelerate the training process such as :(1) early stop, because the best epoch of our BBN is usually not the final one, (2) concat the two batches of data, feed them into the shared network and then split them. Hope that will be useful for you !

Kouuh commented 3 years ago

When I use BBN for other data sets, how do I determine the total number of epochs and LR_STEP