DN_CaffeScript
DN_CaffeScript copied to clipboard
Tongcheng
Hi, Tongcheng. Can you figure out how to integrate the DenseBlock_layer into weiliu89's caffe SSD branch? As my understanding, you provide the 4 important files: DenseBlock_Layer.hpp, DenseBlock_Layer.cpp, DenseBlock_Layer.cu, test_DenseBlock_Layer.cpp to...
Hi Tongcheng, Thank you for your implementation of Densenet Caffe and I can already train Cifar 10 dataset. However, when it comes to my own dataset, it will be stuck...
I tried to convert pretrained densenet121 model provided in https://github.com/shicai/DenseNet-Caffe to efficient version obeying your DenseBlock naming convention. I have the following prototxt (efficient_densenet121.prototxt) and script to copy params (from...
I have carefully read the authors' original paper. they mention their DenseNet-BC as a composition of 'DenseNet-B' and 'DenseNet-C' (see paper's sec-3 'Bottleneck layers' & 'Compression'). In 'DenseNet-B', a hyper-parameter...
i was going to pick up the training process with .solverstate file, but run into the following problem. ``` F0816 17:25:53.431728 27270 layer_factory.hpp:81] Check failed: registry.count(type) == 1 (0 vs....
When i run this caffeSript, i found the speed improved about 3 times than original caffe(maybe the new net is small?) , which from 60ms to 20ms in GTX1080, but...
I did find pretrained models of imagenet dataset with a normal vanilla implementation(https://github.com/shicai/DenseNet-Caffe) for densenet. But since you have changed the names of layer(now a denseblock layer), that pretrained wont...
This [version](https://github.com/Tongcheng/caffe/) of caffe with cmake is unable to find Cuda, hdf5, bunch of other things and even python on my system. Strange thing is, I have earlier compiled Faster-RCNN...
I found that CuDNNBatchNormLayer in your caffe branch ( https://github.com/Tongcheng/caffe ) is not registered as a Creator in layer_factory.cpp, But your BatchNorm Registered with REGISTER_LAYER_CLASS(BatchNorm); An registered CuDNNBatchNormLayer seems like...
