Leo Chenghui Li
Leo Chenghui Li
Hi @shicai I only got 54 as top 1 acc and 79 as top 5 acc. Btw I didn't use the pretrained model, I'm trying to train the net from...
@shicai Hi shicai, I've tuned the top 1 acc to 67%, still sort of a long way from the paper acc. I trained with batch size 256, 90 epochs and...
PyTorch model seems not working so well? FYI, I am using PyTorch backend and your pretrained model. Here's the output of your own test image: 
Hi Richard, I was using the pytorch_trained.pth fetched from fetch_models.sh
Hi @wenwei202 , I have enabled MKL for BLAS before this error. What else do you think could be wrong? Bests
Hi @wenwei202 , As I cannot test the performance of sparsified convolutions, it would be very nice of you to share the sparse scheme timing Something similar to: > I0907...
Hi @farmingyard , I'm sort of confused about your stride case in your deploy prototxt. Why you don't shuffle the channel here? thanks!
@farmingyard thanks man! Btw what's your acc for this? I only got 54% as top1_acc and 79% as top5_acc. According to the paper it's only around 34.1% error rate. I...
Hi @farmingyard , Do you finally reach the 65.9% top 1 acc in the paper? I trained with: batchsize 256, totally 100 epochs, base lr: 0.1 decay the learning rate...
@xiaomr Hi, I'm not sure tho. Since the depthwise conv layer are not designed for all parallel-GPU systems, if you have your own parallel GPU system, you may need to...