caffe

问题 I'm trying to train a LeNet on my own data (37 by 37 grayscale images of 1024 categories). I created the lmdb files, and changed the size of the ouput layer to 1024. When I ran the caffe train with my solver file, the program got stuck after printing ... layer { name: "loss" type: "SoftmaxWithLoss" bottom: "score" bottom: "label" top: "loss" } I0713 17:11:13.334890 9595 layer_factory.hpp:77] Creating layer data I0713 17:11:13.334939 9595 net.cpp:91] Creating Layer data I0713 17:11:13.334950

问题 i have a question regarding Image preprocessing in Caffe. When i use the BatchNormalization Layer in my caffemodel, do i still need the preprocessing step "image mean subtraction" on all my trainings Images before Training Phase starts? Or is this done in the BatchNormalization Layer ? Thank you very much =) 回答1: Image mean subtraction does something different than BatchNormalization and is used for a different purpose. BatchNormalization normalizes a batch and not every single image and is

问题 Looking at the code of "Accuracy" layer, I see there is an option for a second output/"top" for this layer. What does this second output produce? 回答1: Looking at accuracy_layer.hpp, where the number of outputs for the layer are defined, there's this comment: // If there are two top blobs, then the second blob will contain // accuracies per class. So, the second "top" of the "Accuracy" layer simply reports per-class accuracies. Just as a side note for layer Accuracy, the reported Accuracy is

问题 I want to use VGG_ILSVRC_19_layers as a pretrained model in digits but with different dataset. Do I need different label files? How can I upload this model and use it for my dataset? for the VGG 16 layers I got ERROR: Cannot copy param 0 weights from layer 'fc6'; shape mismatch. Source param shape is 1 1 4096 25088 (102760448); target param shape is 4096 32768 (134217728). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer. how can modify

问题 I want to use HDF5 file to input my data and labels in my CNN. I created the hdf5 file with matlab. Here is my code: h5create(['uNetDataSet.h5'],'/home/alexandra/Documents/my-u-net/warwick_dataset/Warwick_Dataset/train/image',[522 775 3 numFrames]); h5create(['uNetDataSet.h5'],'/home/alexandra/Documents/my-u-net/warwick_dataset/Warwick_Dataset/train/anno',[522 775 3 numFrames]); h5create(['uNetDataSet.h5'],'/home/alexandra/Documents/my-u-net/warwick_dataset/Warwick_Dataset/label',[1 numFrames

问题 I am using bvlc_reference_caffenet for training. I am doing both training and testing. Below is an example log of my trained network: I0430 11:49:08.408740 23343 data_layer.cpp:73] Restarting data prefetching from start. I0430 11:49:21.221074 23343 data_layer.cpp:73] Restarting data prefetching from start. I0430 11:49:34.038710 23343 data_layer.cpp:73] Restarting data prefetching from start. I0430 11:49:46.816813 23343 data_layer.cpp:73] Restarting data prefetching from start. I0430 11:49:56

问题 I would like to check if my understanding in organizing my folders and labeling is correct regarding Caffe's way of doing it. My train directory structure looks like below: ~/Documents/software_dev/caffe/data/smalloffice/images/train a_person not_a_person train.txt where both a_person and not_a_person are directories. My train.txt file looks like below: train.txt: ---------- not_a_person/1_rotated.jpg 0 not_a_person/2_rotated.jpg 0 not_a_person/3_rotated.jpg 0 not_a_person/4_rotated.jpg 0 not

问题 Assuming we have a layer like this: layer { name: "fully-connected" type: "InnerProduct" bottom: "bottom" top: "top" inner_product_param { num_output: 1 } } The output is batch_size x 1. In several papers (for exmaple link1 page 3 picture on the top, or link2 page 4 on top)I have seen that they used such a layer in the end to come up with a 2D image for pixel-wise prediction. How is it possible to transform this into a 2D image? I was thinking of reshape or deconvolution, but I cannot figure

问题 I have a volumetric data with size 256x128x256 . Due to limit memory, I cannot use the whole data to directly feed to CAFFE. Hence, I will randomly choose n_sample patches 50x50x50 which extract from the volumetric data and store them into HDF5. I was successful to randomly extract the patches from raw data and its label by extract_patch_from_volumetric_data function. I want to store these patches into the HDF5 data. The bellow code performs the task. Could you look at and verify help me my

问题 I have a network. In one place I want to use concat. As on this picture. Unfortunately, the network doesn't train. To understand why I want to change weights in concat. Meaning that all values from FC4096 will get 1 and all values from FC16000 will get 0 at the beginning. I know that FC4096 will get me 57% accuracy, so with learning rate 10^-6 I will understand why after concatenation layers didn't learn. The question is, how can I set all values from FC4096 to 1 and all values from FC16000