Yunpeng

Hi @VectorYoung , 1. Yes, it uses random sampling. 2. It means the data loader cannot correct extract "frame 30" from that ".avi" file. It is either caused by the...

There is no plan to train DPNs on Caffe. But you can find some converted model at: [soeaver/caffe-model](https://github.com/soeaver/caffe-model/tree/master/cls) Also, you can convert the trained mxnet model by using model converters:...

@whcacademy Please set the `unchanged` option to `1` while generating the `.rec` file using [im2rec](https://github.com/cypw/mxnet/blob/master/tools/im2rec.cc) as I have already indicated in [run_val.sh](https://github.com/cypw/DPNs/blob/master/run_val.sh#L4). Note that: By setting `unchanged=1`, `im2rec` will directly...

@yushanshan05 @whcacademy You should be able to fully reproduce the result by the following steps: ( I've double checked by myself ) 1. Download ImageNet-1k validation set 2. `git clone...

@rxdhr I would suggest you check if the pre-trained model is correctly loaded, and check if your fine-tuning code/strategy* is correct: This can be done by fine-tuning the pre-trained model...

Hmmm, that's an interesting observation. 32 images per GPU should be fine. Maybe fine-tuning DPN-107 requires smaller initial learning rate than ResNet/ResNeXt since it has an additional Dense Path? In...

The training steps is shown [here](https://github.com/cypw/PyTorch-MFNet/blob/master/train_kinetics.py#L61) and [here](https://github.com/cypw/PyTorch-MFNet/blob/master/train_model.py#L101), i.e. about 100 epochs. I set the `--end-epoch=10000` just because I want to manually terminate the training when validation accuracy saturated.

[mxnet/src/operator/matrix_op-inl.h \#L843](https://github.com/cypw/mxnet/blob/92053bd3e71f687b5315b8412a6ac65eb0cc32d5/src/operator/matrix_op-inl.h#L843) ```c++ // slice_axis MXNET_REGISTER_SIMPLE_OP(slice_axis, XPU) .set_enable_kwargs(true) .set_function(XPU::kDevMask, Slice, kNoInplace, kRegisterSymbolic) .set_gradient(XPU::kDevMask, SliceGrad_, kNoInplace) .set_shape_function(SliceShape) .describe("Slice the input along certain axis and return a sliced array.") .add_arguments(SliceParam::__FIELDS__()); ```

@LeonJWH MXNet does not store information about "moving_mean" and "moving_variance" in the json file. ( see: http://data.dmlc.ml/mxnet/models/imagenet/resnet ) Please ask this question at MXNet repo for more information. Thanks!

@LeonJWH No, we didn't merge them into any other params. ( see: [forward code](https://github.com/cypw/mxnet/blob/92053bd3e71f687b5315b8412a6ac65eb0cc32d5/src/operator/batch_norm-inl.h#L109) ) You can get these raw values by `_, _, aux_params = mx.model.load_checkpoint(prefix, epoch)` ( see:...