pistony / modelzoo.pytorch Goto Github PK

Hi, could you please share the pretrained weights of LamdaNetworks on Imagenet?

Hi there, while browsing the official GhostNet repository on Gitee.com, I found a potentially very interesting version of GhostNet termed GhostNet-600 which claims 80.2 top1 accuracy on ImageNet for only 0.6GFLOPS and 11.9M parameters. It is not mentioned in the paper but the source code is available on Gitee.
Could you reproduce this performance and provide latency information ?

Hi,

Thanks for sharing this great work. I try to run you command to reproduce ghostnet but I'm only able to get around 74.7%, which should be >75%. Is there any suggestions? Here is the command I use:

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python distribute_train_script.py --data-path /workspace/datasets/ImageNet/decompressed/t1/ --batch-size 256 --dtype float16 \
                                  -j 48 --epochs 360 --lr 2.6 --warmup-epochs 5 --label-smoothing \
                                  --no-wd --wd 0.00003 --model GhostNet --log-interval 150 --model-info \
                                  --dist-url tcp://127.0.0.1:26548 --world-size 1 --rank 0

Thanks for your great work!
Could you please share the influence of the batch size and the number of GPUs?
Also how to choose a suitable learning rate and batch size if the available GPUs is not enough.
Thank you!