Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update Regnet model weights #4530

Merged
merged 5 commits into from
Oct 5, 2021
Merged

Update Regnet model weights #4530

merged 5 commits into from
Oct 5, 2021

Conversation

kazhang
Copy link
Contributor

@kazhang kazhang commented Oct 4, 2021

When I uploaded the RegNet model weights, I picked the last checkpoint rather than the checkpoints that generated best Acc@1. (for regnet_y_800mf and regnet_x_400mf, the best checkpoints are the last checkpoints, so they are not updated.)

In addition, regnet_y_1_6gf, regnet_y_3_2gf, regnet_y_16gf, regnet_y_32gf, regnet_x_1_6gf, regnet_x_3_2gf, regnet_x_16gf, regnet_x_32gf are also included. We have weights for all variants now.

Resolves #2655

cc @datumbox

@kazhang
Copy link
Contributor Author

kazhang commented Oct 4, 2021

Tested pretrained weights on ImageNet.

updated weights

model Acc@1 Paper Acc@1 Acc@5 command
regnet_y_400mf 74.046 74.1 91.716 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_y_400mf --batch-size 128 --test-only --pretrained
regnet_y_8gf 80.032 79.9 95.048 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_y_8gf --batch-size 64 --test-only --pretrained
regnet_x_800mf 75.212 75.2 92.348 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_x_800mf --batch-size 128 --test-only --pretrained
regnet_x_8gf 79.344 79.3 94.686 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_x_8gf --batch-size 64 --test-only --pretrained

new weights

model Acc@1 Paper Acc@1 Acc@5 command
regnet_y_1_6gf 77.95 78 93.966 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_y_1_6gf --batch-size 128 --test-only --pretrained
regnet_y_3_2gf 78.948 79 94.576 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_y_3_2gf --batch-size 64 --test-only --pretrained
regnet_y_16gf 80.424 80.4 95.24 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_y_16gf --batch-size 32 --test-only --pretrained
regnet_y_32gf 80.878 81 95.34 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_y_32gf --batch-size 32 --test-only --pretrained
regnet_x_1_6gf 77.04 77 93.44 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_x_1_6gf --batch-size 128 --test-only --pretrained
regnet_x_3_2gf 78.364 78.3 93.992 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_x_3_2gf --batch-size 64 --test-only --pretrained
regnet_x_16gf 80.058 80 94.944 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_x_16gf --batch-size 64 --test-only --pretrained
regnet_x_32gf 80.622 80.5 95.248 srun -p $TRAIN_QUEUE --cpus-per-task=96 --gpus-per-node=8 python -m torch.distributed.launch --nproc_per_node=8 --use_env train.py --model regnet_x_32gf --batch-size 32 --test-only --pretrained

@kazhang kazhang requested a review from datumbox October 4, 2021 07:20
@kazhang kazhang marked this pull request as ready for review October 4, 2021 07:20
Copy link
Contributor

@datumbox datumbox left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @kazhang, looks good to me. Merge when ready.

@kazhang kazhang merged commit 0c0a6a4 into pytorch:main Oct 5, 2021
@kazhang kazhang deleted the models/regnet-weights branch October 5, 2021 04:46
kazhang added a commit that referenced this pull request Oct 6, 2021
* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add weights for x_16gf, x_32gf, y_16gf
facebook-github-bot pushed a commit that referenced this pull request Oct 7, 2021
Summary:
* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add weights for x_16gf, x_32gf, y_16gf

Reviewed By: kazhang

Differential Revision: D31387949

fbshipit-source-id: ae8c6d5c2a391cc25e441648d57d304cc89b5646
mszhanyi pushed a commit to mszhanyi/vision that referenced this pull request Oct 19, 2021
* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add weights for x_16gf, x_32gf, y_16gf
cyyever pushed a commit to cyyever/vision that referenced this pull request Nov 16, 2021
* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add best weights and x_1_6, x_3_2, y_1_6, y_3_2, y_32 weights

* add weights for x_16gf, x_32gf, y_16gf
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

RegNet in torchvision ?
3 participants