On-the-fly net resizing, without reallocation (where possible)#594

Merged

shelhamer merged 19 commits intoBVLC:devfrom longjon:layer-reshaping

Sep 18, 2014

+575-210

Commits on Sep 18, 2014

use Blob directly instead of shared_ptr for EltwiseLayer::max_idx_

longjon
authored and
shelhamer
committed
add abstract Layer::Reshape, and document the new method protocol

longjon
authored and
shelhamer
committed
don't reallocate blobs when shrinking memory use

longjon
authored and
shelhamer
committed
enable reshaping in the forward pass

longjon
authored and
shelhamer
committed
separate setTensor4dDesc from createTensor4dDesc

longjon
authored and
shelhamer
committed
separate setConvolutionDesc from createConvolutionDesc

longjon
authored and
shelhamer
committed
split off Reshape for data layers

longjon
authored and
shelhamer
committed
split off Reshape for loss layers

longjon
authored and
shelhamer
committed
split off Reshape for neuron layers

longjon
authored and
shelhamer
committed
split off Reshape for common layers

longjon
authored and
shelhamer
committed
split off Reshape for vision layers

longjon
authored and
shelhamer
committed
call Reshape in Layer::SetUp

longjon
authored and
shelhamer
committed
default LayerSetUp to no-op instead of NOT_IMPLEMENTED

longjon
authored and
shelhamer
committed
test net reshaping

longjon
authored and
shelhamer
committed
include Reshape in caffe time

longjon
authored and
shelhamer
committed
add Net::Reshape for only reshaping

longjon
authored and
shelhamer
committed
[pycaffe] expose Net::Reshape

longjon
authored and
shelhamer
committed
[docs] clarify the use of Blob::Reshape a bit

longjon
authored and
shelhamer
committed
check that LRN's local_size is odd as the current implementation requires

longjon
authored and
shelhamer
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

On-the-fly net resizing, without reallocation (where possible)#594

Commits on Sep 18, 2014

use Blob directly instead of shared_ptr for EltwiseLayer::max_idx_

add abstract Layer::Reshape, and document the new method protocol

don't reallocate blobs when shrinking memory use

enable reshaping in the forward pass

separate setTensor4dDesc from createTensor4dDesc

separate setConvolutionDesc from createConvolutionDesc

split off Reshape for data layers

split off Reshape for loss layers

split off Reshape for neuron layers

split off Reshape for common layers

split off Reshape for vision layers

call Reshape in Layer::SetUp

default LayerSetUp to no-op instead of NOT_IMPLEMENTED

test net reshaping

include Reshape in caffe time

add Net::Reshape for only reshaping

[pycaffe] expose Net::Reshape

[docs] clarify the use of Blob::Reshape a bit

check that LRN's local_size is odd as the current implementation requires