Cifar10 vgg19 zero remaining channel (network slimming) about rethinking-network-pruning

[Eric-mingjie]

Did you apply sparsity during training for vgg19?

[ivllm]

No, I trained the general baseline model with the command below.
python main.py --dataset cifar10 --arch vgg --depth 19

[Eric-mingjie]

You should apply sparsity during training https://github.com/Eric-mingjie/network-slimming#train-with-sparsity.

[ivllm]

Why the general model has a problem and should use the sparsity model?
Please give me a brief explanation.

[Eric-mingjie]

Yes, this is why network slimming is proposed, general model can have this issue of 0 channels after pruning. You can read the original paper https://arxiv.org/abs/1708.06519.

[ivllm]

I got it. Thank you for your kind reply