59f4eefb7f
Merge branch 'attention' into EcaBam
6 years ago
fa38f24967
Update SK network configs, add weights for skresnet8 and skresnext50
6 years ago
5d2df0746e
refactor cecabam with cbam
6 years ago
1448a19d67
Update .gitignore
...
add testing notebook into gitignore
6 years ago
26922eb2ce
Merge branch 'attention' into EcaBam
6 years ago
a4b1993c84
Restore Bias
...
The bias term might be useful for actual spatial 2d convs
6 years ago
5e6dbbaf30
Add CBAM for experimentation
6 years ago
d725991870
Remove debug print from ECA module
6 years ago
2a7d256fd5
Re-enable mem-efficient/jit activations after torchscript tests
6 years ago
f902bcd54c
Layer refactoring continues, ResNet downsample rewrite for proper dilation in 3x3 and avg_pool cases
...
* select_conv2d -> create_conv2d
* added create_attn to create attention module from string/bool/module
* factor padding helpers into own file, use in both conv2d_same and avg_pool2d_same
* add some more test eca resnet variants
* minor tweaks, naming, comments, consistency
6 years ago
a99ec4e7d1
A bunch more layer reorg, splitting many layers into own files. Improve torchscript compatibility.
6 years ago
13746a33fc
Big move, layer modules and fn to timm/models/layers
6 years ago
f54612f648
Merge branch 'select_kernel' into attention
6 years ago
4defbbbaa8
Fix module name mistake, start layers sub-package
6 years ago
7011cd0902
A little bit of ECA cleanup
6 years ago
46471df7b2
Merge pull request #82 from VRandme/eca
...
ECA-Net Efficient Channel Attention
6 years ago
d0eb59ef46
Remove unused default_init for EfficientNets, experimenting with fanout calc for #84
6 years ago
cade829105
Add EfficientNet-ES to sotabench
6 years ago
dbdb3ce963
Add CecaBam
...
CecaBam is circular ECA followed by BAM for both channelwise and spatial attention
6 years ago
e6a762346a
Implement Adaptive Kernel selection
...
When channel size is given,
calculate adaptive kernel size according to original paper.
Otherwise use the given kernel size(k_size), which defaults to 3
6 years ago
13e8da2b46
SelectKernel split_input works best when input channels split like grouped conv, but output is full width. Disable zero_init for SK nets, seems a bad combo.
6 years ago
6db087a1ff
Merge remote-tracking branch 'upstream/master' into eca
6 years ago
904c618040
Update EcaModule.py
...
Make pylint happy
(commas, unused imports, missed imports)
6 years ago
db91ba053b
EcaModule(CamelCase)
...
CamelCased EcaModule.
Renamed all instances of ecalayer to EcaModule.
eca_module.py->EcaModule.py
6 years ago
5eb0e363a6
Update README.md
6 years ago
5c4991a088
Add PyTorch trained EfficientNet-ES weights from Andrew Lavin
6 years ago
d04ff95eda
Merge branch 'master' into eca
6 years ago
d63ae121d5
Clean up eca_module code
...
functionally similar
adjusted rwightman's version of reshaping and viewing.
Use F.pad for circular eca version for cleaner code
6 years ago
b0d45fd09c
Merge remote-tracking branch 'upstream/master'
6 years ago
d66819d1f3
Indentation mistake. Fixes #81
6 years ago
65cda1c63f
Merge pull request #83 from andravin/validation-batch-size-multiplier
...
Added commandline argument validation-batch-size-multiplier.
6 years ago
f87fcd7e88
Implement Eca modules
...
implement ECA module by
1. adopting original eca_module.py into models folder
2. adding use_eca layer besides every instance of SE layer
6 years ago
697e05cb3e
Merge remote-tracking branch 'upstream/master'
6 years ago
b72013def8
Added commandline argument validation-batch-size-multiplier with default set to 1.
6 years ago
4808b3c32f
Bump version for PyPi update, fix few out of date README items/mistakes, add README updates for TF EfficientNet-B8 (RandAugment)
6 years ago
5c85389e26
Update README.md
6 years ago
820b73d280
Update README.md
6 years ago
82c0a2ff12
Update README.md
...
Fix relative paths (I think)
6 years ago
1ffd2d0694
Add results/README.md
6 years ago
89a32454e1
Merge remote-tracking branch 'upstream/master'
6 years ago
7d07ebb660
Adding some configs to sknet, incl ResNet50 variants from 'Compounding ... Assembled Techniques' paper and original SKNet50
6 years ago
a9d2424fd1
Add separate zero_init_last_bn function to support more block variety without a mess
6 years ago
355aa152d5
Just leave it float for now, will look at fp16 later. Remove unused reference code.
6 years ago
ef457555d3
BlockDrop working on GPU
6 years ago
3ff19079f9
Missed nn_ops.py from last commit
6 years ago
9f11b4e8a2
Add ConvBnAct layer to parallel integrated SelectKernelConv, add support for DropPath and DropBlock to ResNet base and SK blocks
6 years ago
cefc9b7761
Move SelectKernelConv to conv2d_layers and more
...
* always apply attention in SelectKernelConv, leave MixedConv for no attention alternative
* make MixedConv torchscript compatible
* refactor first/previous dilation name to make more sense in ResNet* networks
6 years ago
9abe610931
Used wrong channel var for split
6 years ago
58e28dc7e7
Move Selective Kernel blocks/convs to their own sknet.py file
6 years ago
a93bae6dc5
A SelectiveKernelBasicBlock for more experiments
6 years ago
Chris Ha