Proto commits in gustavla/caffe-weighted-samples

These commits are when the Protocol Buffers files have changed: (only the last 100 relevant commits are shown)

2014-11-05

Added SOFTMAX_CROSS_ENTROPY_LOSS layer. This allows training with soft targets.

The documentation is generated from this commit.

2014-11-04

Added the essentials to use weighted samples. - Change data layers to read sample_weight. - Modified SOFTMAX_LOSS and EUCLIDEAN_LOSS to understand sample_weight.

2014-10-16

Merge pull request #190 from sguada/new_lr_policies New lr policies, MultiStep and StepEarly

Added cache_images to WindowDataLayer Added root_folder to WindowDataLayer to locate images

Add root_folder to ImageDataLayer

2014-10-15

Added encoded datum to io

2014-10-14

Refactored leveldb and lmdb code.

2014-10-10

Added Multistep, Poly and Sigmoid learning rate decay policies Conflicts: include/caffe/solver.hpp src/caffe/proto/caffe.proto src/caffe/solver.cpp

2014-10-06

correct naming in comment and message about average_loss

2014-10-04

Merge pull request #1070 from sguada/move_data_mean Refactor data_transform to allow datum, cv:Mat and Blob transformation

2014-10-03

Added mean_value to specify mean channel substraction Added example of use to models/bvlc_reference_caffenet/train_val_mean_value.prototxt

Added global_pooling to set the kernel size equal to the bottom size Added check for padding and stride with global_pooling

Initial cv::Mat transformation Added cv::Mat transformation to ImageDataLayer Conflicts: src/caffe/layers/image_data_layer.cpp Added transform Datum to Blob Conflicts: src/caffe/layers/base_data_layer.cpp src/caffe/layers/base_data_layer.cu Added transform cv::Mat to Blob Added transform Vector<Datum> to Blob Conflicts: src/caffe/data_transformer.cpp

2014-10-02

Add ExpLayer to calculate y = base ^ (scale * x + shift)

caffe.proto: do some minor cleanup (fix comments, alphabetization)

2014-10-01

Merge pull request #1096 from qipeng/smoothed-cost Display averaged loss over the last several iterations

2014-09-22

added support for "k" LRN parameter to upgrade_proto

adds a parameter to the LRN layer (denoted as "k" in [Krizhevsky et al., NIPS 2012])

2014-09-19

Merge pull request #959 from nickcarlevaris/contrastive_loss Add contrastive loss layer, tests, and a siamese network example

2014-09-17

Display averaged loss over the last several iterations

2014-09-14

fix spelling error in caffe.proto

fix out-of-date next ID comment for SolverParameter

2014-09-08

Added contrastive loss layer, associated tests, and a siamese network example using shared weights and the contrastive loss.

Implemented elementwise max layer

2014-09-07

add cuDNN to build

add engine parameter for multiple computational strategies add `engine` switch to layers for selecting a computational backend when there is a choice. Currently the standard Caffe implementation is the only backend.

groom proto: sort layer type parameters, put loss_weight after basics

2014-09-05

Merge pull request #1036 from longjon/test-initialization-param Add test_initialization option to allow skipping initial test

2014-09-04

add SILENCE layer -- takes one or more inputs and produces no output This is useful for suppressing undesired outputs.

add test_initialization option to allow skipping initial test

2014-09-03

Move transform param one level up in the proto to reduce redundancy

Add transformer to the memory data layer

2014-09-01

Added L1 regularization support for the weights

cleanup caffe.proto

proto conflit, lint, and math_functions (compiler complaint)

fixes after rebase

fixed caffe.proto after a mistaken rebase

2014-08-25

Add "stable_prod_grad" option (on by default) to ELTWISE layer to compute the eltwise product gradient using a slower but stabler formula.

2014-08-22

restore old data transformation parameters for compatibility

2014-08-21

Refactor ImageDataLayer to use DataTransformer

2014-08-20

Refactor DataLayer using a new DataTransformer Start the refactoring of the datalayers to avoid data transformation code duplication. So far, only DataLayer has been done.

2014-08-17

cmake build system

2014-08-15

Add "not_stage" to NetStateRule to exclude NetStates with certain stages.

2014-08-14

Added absolute value layer, useful for implementation of siamese networks! This commit also replaces the default caffe_fabs with MKL/non-MKL implementation of Abs.

2014-08-13

Add loss_weight to proto, specifying coefficients for each top blob in the objective function.

2014-08-11

added cross-channel MVN, Mean-only normalization, added to layer factory, moved to common_layers

mean-variance normalization layer

2014-08-07

LICENSE governs the whole project so strip file headers

2014-08-04

Add 'snapshot_after_train' to SolverParameter to override the final snapshot.

2014-07-29

Add NetState message with phase, level, stage; NetStateRule message with filtering rules for Layers.

add h/w kernel size, stride, and pad for non-square filtering while keeping everything working as-is.

2014-07-26

Print blob L1 norms during forward/backward passes and updates if new "debug_info" field in SolverParameter is set.

2014-07-22

SliceLayer: post-rebase fixes, cleanup, etc. (some from changes suggested by @sguada). Test for both num & channels in forward & backward; use GaussianFiller so that tests are non-trivial.

Add split dim layer Differentiate top test blob vector size Rename to SplitLayer Add slicing points

2014-07-21

leaky relu + unit test

2014-07-19

Extend the ArgMaxLayer to output top k results

2014-07-18

cosmetics: add syntax = proto2

2014-07-05

add tests for rectangular pooling regions

2014-07-04

Update caffe.proto Add pad_h, pad_w, kernel_size_h, kernel_size_w, stride_h, stride_w to support pooling on rectangle regions.

2014-06-27

Next LayerParameter proto id

Update name of last added param.

Incorporate top_k param into AccuracyLayer and check it's value.

Add parameter for AccuracyLayer in proto.

2014-06-26

change weight blob field name to param

weight sharing

2014-06-21

Remove C_ mentions, extra spaces and change hinge_norm to norm

Removed L2HingeLoss class now a case within HingeLoss class Conflicts: include/caffe/vision_layers.hpp src/caffe/layers/loss_layer.cpp src/caffe/proto/caffe.proto src/caffe/test/test_l2_hinge_loss_layer.cpp

Merge HingeLoss and L2HingeLoss by adding hinge_norm to params Conflicts: src/caffe/layers/loss_layer.cpp src/caffe/proto/caffe.proto src/caffe/test/test_l2_hinge_loss_layer.cpp

2014-06-13

add option for lmdb

2014-06-09

add DummyDataLayer

2014-06-08

padding for max pooling Max pooling pads by -inf if the padding parameter is set. Padding for pooling, like padding for convolution, can preserve the dimensions of the bottom at the top. By setting the padding to floor(kernel_size / 2) the top output is the "same" instead of the "valid" part of the bottom input.

2014-05-27

Fixed ThresholdParam Conflicts: src/caffe/proto/caffe.proto Conflicts: src/caffe/proto/caffe.proto Conflicts: src/caffe/proto/caffe.proto

2014-05-23

commment, lint

weight elementwise sum with per-blob coefficients

make sum the default eltwise operation

2014-05-22

Elementwise layer learns summation

EltwiseProductLayer -> EltwiseLayer for generality Reproduce elementwise product layer in more generality. Add elementwise operation parameter. Prepare for elementwise sum operation choice.

2014-05-21

corrected the caffe.proto ids

Added ArgMax Layer Conflicts: src/caffe/proto/caffe.proto

Fixed numbers in proto and name of ArgMaxParameter Conflicts: src/caffe/proto/caffe.proto

Fix types of ArgMax Layers params Conflicts: include/caffe/vision_layers.hpp src/caffe/proto/caffe.proto

2014-05-10

fix proto comment for multiple test nets

multiple test_iter

allow multiple test nets

specify NetParameters directly in the SolverParameter

2014-05-09

make solver_mode an enum with CPU and GPU -- fully backwards compatible with old 0/1 style

2014-05-02

add MemoryDataLayer for reading input from contiguous blocks of memory

2014-04-30

note the last added layer/params in caffe.proto to prevent conflicts The current scheme does not actually prevent conflicts, since three-way merge will accept simultaneous changes that agree on the next number. This commit fixes this by explicitly noting the last layer added.

2014-04-25

add HingeLossLayer for one-vs-all hinge loss This layer implements a "one-vs-all" hinge loss, (1/n) sum_ij max(0, 1 - y_ij x_ij), with bottom blob x_ij (i ranging over examples and j over classes), and y_ij = +1/-1 indicating the label. No regularization is included, since regularization is done via weight decay or using the parameters of another layer. The gradient is taken to be zero at the hinge point. This commit only provides the CPU implementation.

2014-04-23

add random_seed field to SolverParameter and have solver use it -- already works for lenet, doesn't work for imagenet w/ rand() calls

2014-04-16

change to correct next layer id for merge

2014-04-15

make solver able to compute and display test loss

add mnist autoencoder example necessities (sigmoid cross entropy loss layer, sparse gaussian filler)

2014-04-08

update proto field IDs from placeholder values

merge LRNMapLayer into LRNLayer with norm_region proto field

add LRN within map layer and dependencies (eltwise product and power)

use average pool instead of conv

minor polishing

add padding for average pooling

2014-03-28

some post rebase fixes -- copyright, hdf5_output layer (still need to incorporate into util/upgrade_proto)