Yunpeng

Hi @xioryu , I also cannot reproduce the reported accuracy. I doubt if the reported results are reliable since I noticed that the reported accuracy is posted before the evaluation...

32 GB RAM on the CPU side should be enough. You can try to reduce `num_woker` to save some RAM: https://github.com/cypw/PyTorch-MFNet/blob/master/data/iterator_factory.py#L183

``` % Matlab Code % FLOPs for nn.Conv3d, without bias A_size = [num_out/group, prod(kernel)*num_input/group]; % weight C_size = [prod(kernel)*num_input/group, out_h*out_w*out_t]; % im2col flops = 0; for i_group = 1:group flops...

Hi @BigFishMaster , Thanks for pointing out my mistake. It seems that I have shared the wrong train/val list for the 5k models. (I might have removed some categories from...

I haven't found any HLS implementation on GPU. It might be helpful if the color augmentation could be done on the GPU side. Besides considering reducing the cost of data...

The provided MXNet in the DPN repo is a pretty old version which may not be compatible with `.rec` generated by the newer version. You can either generate the `.rec`...

Have you followed my recommendations above? There are numerous mistakes that lead to `Segmentation fault`. Could you provide me with more details?

@shipeng-uestc If you are using the provided old [MXNet \@ 92053bd](https://github.com/cypw/mxnet/tree/92053bd3e71f687b5315b8412a6ac65eb0cc32d5), please do the following tests to debug: Step 1: Make sure your `train.rec` and `val.rec` are correct and your...

I fix it to `0.765625` for evaluating a 224x224 center crop with all input images resized to `short length = 256`. Note: `min_random_area = max_random_area = 0.765625 = (224^2)/(256^2)`. Yes,...

@shipeng-uestc The last two `*.rec` files are unnecessary since all resizing can be done inside the data iterator. Actually, I don't quite understand why do you need to use the...