Skip to content

visualization of the conv layer kernels #158

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Closed
wants to merge 40 commits into from
Closed

visualization of the conv layer kernels #158

wants to merge 40 commits into from

Conversation

tdomhan
Copy link
Contributor

@tdomhan tdomhan commented Feb 25, 2014

This is basically an adopted version of the conv layer kernel visualization from: https://code.google.com/p/cuda-convnet/source/browse/trunk/shownet.py

Here's for example the first layer of the imagenet model, that's available for download:
out

Aravindh Mahendran and others added 30 commits February 18, 2014 00:49
Added tanh activation function layer.
68f51c8
Added a test for the tanh layer.
9a45a0a
@jeffdonahue @shelhamer
add SplitLayer and Net::AddSplits to transform shared bottom blobs into
207fd71 
split layers
@jeffdonahue @shelhamer
make split_layer backward obey propagate_down
08bbf6a
@jeffdonahue @shelhamer
add split layer tests
36db05f
@jeffdonahue @shelhamer
add split layer tests
b2a78f6
@jeffdonahue @shelhamer
add split layer insertion tests; move split insertion code to util file
0082ac4
@jeffdonahue @shelhamer
some cleanup
9452a31
@shelhamer
Merge pull request BVLC#116 from aravindhm/tanh
2fce080 
Add TanH = hyperbolic tangent activation layer (popular for sparse
autoencoders).
@jeffdonahue @shelhamer
eliminate redundant code with get_split_blob_name method
ec19814
@jeffdonahue @shelhamer
change \n's to less distracting spaces in hard-coded proto strings
10d7869
@jeffdonahue @shelhamer
give first top split blob same name as bottom blob
26414dd
@jeffdonahue @shelhamer
allow in place computation of SplitLayer 0th top blob
98180ff
@jeffdonahue @shelhamer
change \" in test_split_layer to ' for readability
e948da1
@jeffdonahue @shelhamer
remove pointlessly duplicated CheckGradientExhaustive calls (I screwed
6dc6f1a 
up when merging, I think)
@jeffdonahue @shelhamer
fix comment typo
6e8d332
@jeffdonahue @shelhamer
add test for layer with two tops that are inputs to multiple layers
1800802
@jeffdonahue @shelhamer
remove redundant add_bottom (immediately cleared and then re-added)
f77e25a
@jeffdonahue @shelhamer
get_split_blob_name returns a string to remove some verbosity
03c57f6
@jeffdonahue @shelhamer
get rid of messy snprintf string concatenation
e9228ce
@jeffdonahue @shelhamer
eliminate some cruft by relying on std::map default initializations
fce8f71
@jeffdonahue @shelhamer
remove unnecessary include
2fb43b4
@jeffdonahue @shelhamer
fix split layer insertion bug with in-place layers
f757b6d
@jeffdonahue @shelhamer
add imagenet no split insertion test
bf5e75e
@jeffdonahue @shelhamer
add idempotence test
bb8e0cd
@jeffdonahue @shelhamer
minor cleanup; only get blob_name if needed
26630fe
@shelhamer
Merge pull request BVLC#129 from jeffdonahue/dags-by-split
5792f44 
Generalize architectures to arbitrary DAGs by split layers
@sguada
Set leveldb options.max_open_files = 100. Fix BVLC#13 and BVLC#38
9c5d807
@shelhamer
Merge pull request BVLC#154 from sguada/leveldb_max_open_files
ee00953 
Set leveldb options.max_open_files = 100 and Fix BVLC#13 and BVLC#38
@mavenlin
implemented padding aware im2col and col2im functions
995351d