-
Notifications
You must be signed in to change notification settings - Fork 1.2k
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
Add RepLKNet_B.yml and replknet.py #2786
base: develop
Are you sure you want to change the base?
Conversation
the arch of replknet
Thanks for your contribution! |
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
许多训练超参数没有对齐,建议参考下paddleclas中其他模型的配置,如DeiT、TinyNet等。下面这几个是比较明显的问题,建议再确认对齐一下配置:
- 涉及到resize的地方一般需要设置backend为pil;
- batch_size也没有对齐;
- 使用TimmAutoAugment替换randaugment;
- 官方似乎用了color jitter;
color_jitter 在源码中不触发 由于使用了autoaugment
已经修改数据预处理部分的参数对齐 . batch_size设置为128, update频率设置为8, 总的bs为 12888 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
提交pr的yaml文件中的注释都删掉吧
- CenterCrop: #修改 | ||
size: 224 | ||
- NormalizeImage: | ||
scale: 1.0/255.0 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
这里的格式不对
order: '' | ||
sampler: | ||
name: DistributedBatchSampler | ||
batch_size: 128 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
确认是128吗?我看似乎是64?
python -m torch.distributed.launch --nproc_per_node=8 main.py --model RepLKNet-31B --drop_path 0.5 --batch_size 64 --lr 4e-3 --update_freq 4 --model_ema true --model_ema_eval true --data_path /path/to/imagenet-1k --warmup_epochs 10 --epochs 300 --output_dir your_training_dir
https://github.com/DingXiaoH/RepLKNet-pytorch#pretrain-replknet-31b-on-imagenet-1k
# used for static mode and model export | ||
image_shape: [3, 224, 224] | ||
save_inference_dir: ./inference | ||
update_freq: 8 #修改 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
我看好像是4?
python -m torch.distributed.launch --nproc_per_node=8 main.py --model RepLKNet-31B --drop_path 0.5 --batch_size 64 --lr 4e-3 --update_freq 4 --model_ema true --model_ema_eval true --data_path /path/to/imagenet-1k --warmup_epochs 10 --epochs 300 --output_dir your_training_dir
delete the blank space
No description provided.