Next: release candidate #1112

…ed in convert_mnist_data.cpp to lmdb

…ated description

fix for tools/extra/parse_log.sh

convert MNIST demo to lmdb, fixes

Clean up pycaffe core

…radient and AdaGrad

TestGradientBasedSolver

…ad MNIST example

(euclidean) error

test_iter (should be 100 instead of 50)

convention

better performance) and terminate at iter 65K

enum

Fix pycaffe break due to dropping {blob,layer}.name

…shots

…otxt Snapshot naming issue in mnist_autoencoder_solver_adagrad.prototxt

rules of thumb for SGD

Document Caffe by a tour of key subjects + doxygen dev documentation

Further simplify the data layers and extend the MemoryDataLayer. - design BaseDataLayer and BasePrefectingDataLayer for simplification - refactor DataLayer, ImageDataLayer, and WindowDataLayer - pull up `transform_param` into layer message for de-duplication - add transformation to MemoryDataLayer

This makes the readme file more consistent

Not lenet.prototxt

fix MNIST tutorial details

[example] convert mnist name fix (crashes xcode compiler)

[docs] default setting for layout

… format

This is useful for suppressing undesired outputs.

@longjon

SILENCE layer -- takes 1 or more inputs and produces no output (kudos to @longjon for fixup)

Define standard format for Caffe models to open the "model zoo"

fix gradient-solver test by reducing iterations to relax precision

[example] HDF5 classification

Add test_initialization option to allow skipping initial test

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

Scaffold engine switching for pooling. The Caffe pooling is instantiated without regard for engine in: - LRNLayer - PoolingLayer tests - StochasticPoolingLayer tests - MaxPoolingDropout tests

Add engine switch to pick computational backend

Also, make the parameter name come first. This makes it much easier to find/scan parameters.

It sounds funny to have optional options, and "parameters" is more in line with the internal usage.

Conv layer: fix crash by setting weight pointer

Fixed CMake building test objects multiple times

Fixed CMake script of FindOpenBLAS.

cuDNN acceleration

change uses of tmpnam to mkstemp/mkdtemp

linecount counts more dirs than just src/

compile on OSX w/ CUDA 6.5 This reverts commit 8819f59.

… example using shared weights and the contrastive loss.

@to3i

rebase and fixup #688 from @to3i: elementwise max

Get lmdb from openldap

restore "red X" build failures in Travis

Update CUDA to version 6.5 in the Travis install script

Add ppa for gflag and glog

Previously, the solver constructed nets before the caffe train tool read the --gpu flag, which can cause errors due to LayerSetUp executing on the wrong device (breaking cuDNN, for example).

Fix solver GPU initialization order (e.g., training with cuDNN on non-default device)

…tate

set up datum size for WindowDataLayer

Fix snapshot filename for solver states

Previously, if (height > 0 && width > 0) was true, the cv::resize() function would be called before cv_img_origin was confirmed valid; if the image file/filename was not valid, this caused an opencv assert error like this: terminate called after throwing an instance of 'cv::Exception' what(): /tmp/A3p0_4200_32550/batserve/A3p0/glnxa64/OpenCV/modules/imgproc/src/imgwarp.cpp:1725: error: (-215) ssize.area() > 0 in function resize

[fix] Move file reading error checking closer to actual file read command

…nks.

To thoroughly check convolution, the output is compared against a reference implementation by explicit looping. Simple and group convolution by the Caffe and cuDNN engines are checked against the reference.

Polish mnist + cifar10 examples.

Document and Test Convolution

This is in keeping with #742.

This allows nets to be reshaped very quickly (essentially for free) as long as sufficient memory has been allocated. Calling Blob::Reshape in order to free up memory becomes impossible; however, this is not a normal use case (and deleting blobs does free memory).

Note that calling Reshape when no reshape is necessary should be effectively a no-op, so this is not a performance regression.

This will make it possible to add reshaping to cuDNN layers.

Note that we are dropping some checks from LRN layer. However, these checks are fairly redundant; something is very wrong if these layers are producing top blobs that are different sizes than their inputs, and tests are the right place to catch that. The thing that really should be checked (that isn't) is that that local_size needs to be odd; this will be added in a future commit.

Strictly speaking, Reshape doesn't need to be called until the first Forward call; however, much existing code (especially tests) assumes that top blobs will be set up in SetUp, so we may as well do it there.

Now that top blobs are set up in Layer::Reshape, it's Reshape that is mandatory, and simple layers often don't need to implement LayerSetUp. Reshape is (already) declared abstract, so not implementing it is a compile-time error.

Since we are now calling Reshape in the Forward pass, it's only fair to include it when timing. Reshape calls should normally be four or so orders of magnitude faster than Forward calls; this change also makes it easy to notice a mistake that causes something slow to happen in Reshape.

Note that it is not normally necessary to call this function when using reshapable nets, but sometimes it can be useful to compute the sizes of intermediate layers without waiting for the forward pass.

…ires

On-the-fly net resizing, without reallocation (where possible)

[model zoo] download gist script

- invoke by shell - default download dir to models/ - save to flat dir of owner-gist instead of nested owner/gist

Add contrastive loss layer, tests, and a siamese network example

Commits on Sep 12, 2014

Add ppa for gflag and glog

ste-m5s committed Sep 12, 2014

Configuration menu

View commit details

Copy full SHA for 3a69e22

Browse repository at this point

Copy the full SHA

3a69e22 View commit details

Browse the repository at this point in the history

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Next: release candidate #1112

Next: release candidate #1112

Commits on Aug 31, 2014

Commits on Sep 1, 2014

Commits on Sep 2, 2014

Commits on Sep 3, 2014

Commits on Sep 4, 2014

Commits on Sep 5, 2014

Commits on Sep 6, 2014

Commits on Sep 7, 2014

Commits on Sep 8, 2014

Commits on Sep 10, 2014

Commits on Sep 11, 2014

Commits on Sep 12, 2014

Commits on Sep 13, 2014

Commits on Sep 14, 2014

Commits on Sep 15, 2014

Commits on Sep 16, 2014

Commits on Sep 17, 2014

Commits on Sep 18, 2014

Commits on Sep 19, 2014