This edition of Redstone Torch is published for 2018 Google Science Fair as a snapshot for submission.
The project can be found here: https://www.googlesciencefair.com/projects/2018/f8a2584ecc30ea734d38042a861db599f14dd9a8fb0618b1bb10c480e7dc13b8
python train.py --projecttag base --versiontag base1 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-02-20-23-22-547213-test/ --port=6006
//memory leak 8827 at step 1.3k from 4788
//10 fold 10 train
//BAD MODEL(droped)
=
python train.py --projecttag mem --versiontag mem1 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-02-20-23-22-547213-test/ --port=6006
//memory leak from 8285 at step 1.4k, 6577
//10 fold 1 train
//3*4 epoch, loss=0.5556
=
python train.py --projecttag mem --versiontag mem1 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-03-01-10-04-699788-mem/ --port=6006
//10 fold 10 train, put evaluation inside with each fold instead of epoch
//still memory leak between folds, but leak back when epoch
=
python train.py --projecttag mem2 --versiontag mem2 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-03-02-21-55-372744-mem2/ --port=6006
python predict.py --projecttag 2018-11-03-02-21-55-372744-mem2 --versiontag mem2-pred --loadfile mem2-CP2.pth
//10 fold 2 train, put evaluation back, but save model using self.net and self.optimizers
//memory leak around 7G min stable, 2nd*4 epoch
=
python train.py --projecttag mem3 --versiontag mem3 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-03-14-14-20-075060-mem3/ --port=6006
//No memory leak at epoch 3*4 after delete all extraneous things. memory around 2.5G
//CPU memory leak, GPU fine
=
python train.py --projecttag mem4 --versiontag mem4 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-03-19-02-40-014830-mem4/ --port=6006
python predict.py --projecttag 2018-11-03-19-02-40-014830-mem4 --versiontag mem4 --loadfile mem4-CP9.pth
//open extraneous things, clean-up loss.detach(), clean cache() outside of the epoch(), del more things
//add f1
//no GPU leak during training, but increasing GPU usage after eval
//Epoch: 10*4, Fold: 0 TrainLoss: 0.47 ValidLoss: 0.469516485929, ValidF1: 0.179454994182
//To Resume: python train.py --versiontag 'test' --projecttag 2018-11-04-04-19-26-236033-lr3--loadfile lr3-CP7.pth
//This model is good but it take 15h to get to focal loss 0.5. I guess that is was too small the lr
=
python train.py --projecttag mem5 --versiontag mem5 --resume False (on machine 2)
//add CPU memory monitor and evil monitor
=
python train.py --projecttag lr1 --versiontag lr1 --resume False (on machine 2)
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-04-03-22-36-637908-lr1/ --port=6006
//okay, but not significant lambda x: x/(100*np.mod(-x-1, 600))-0.000006*x
=
python train.py --projecttag lr2 --versiontag lr2 --resume False (on machine 2)
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-04-03-55-40-334831-lr2/ --port=6006
//lambda x: x/(8*np.mod(-x-1, 600)+0.1)-0.000207*x
//ln=2 is fine actually
=
python train.py --projecttag lr3 --versiontag lr3 --resume False (on machine 2)
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-04-04-19-26-236033-lr3/ --port=6006
//adjust batch to 32, start from lr=5
//Epoch: 8*4, Fold: 0 TrainLoss: 0.468069558797 ValidLoss: 0.453331559896, ValidF1: 0.190886673186
=
python train.py --projecttag normal1 --versiontag normal1 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-05-22-05-03-738974-normal1/ --port=6006
python train.py --projecttag 2018-11-05-22-05-03-738974-normal1 --versiontag normal2 --resume True --loadfile normal1-CP1.pth
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-05-22-05-03-738974-normal1/ --port=6006
//normalize data, use both loss(f1, focal), lr=0.1
//lambda global_step: (0.1/2)*(np.cos(np.pi*(np.mod(global_step-1,10000)/(10000)))+1)
//loading speed = 2.46s/it in 32 batch (compare to 1.10s in 16 batch)
//8740k step, 0.6344BestF1, 0.43-45eval-focal, 0.20-21evalF1, 0.2083epochloss, 0.57-58trainF1, 0.44-45trainFocal
=
python train.py --projecttag gpu1 --versiontag gpu1 --resume False (on machine 2)
//around 0.7s per batch of 32
=
python train.py --projecttag tune1 --versiontag tune1 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-07-06-51-09-190794-tune1/ --port=6006
//switch to only one fold, change to Adadelta, adjust lr=2 * 46808 / 32, start lr=1.0, add weighted_bce
//F1 goes up, not good
=
python train.py --projecttag tune2 --versiontag tune2 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-07-07-14-32-212121-tune2/ --port=6006
//log bce, remove bce, maybe at the end of trianing
=
python train.py --projecttag tune3 --versiontag tune3 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-07-07-22-33-042252-tune3/ --port=6006
=
python train.py --projecttag tune4 --versiontag tune4 --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-07-13-02-42-175714-tune4/ --port=6006
python predict.py --projecttag 2018-11-07-13-02-42-175714-tune4 --versiontag tune4 --loadfile tune4-CP11.pth
=
python train.py --projecttag 2018-11-07-13-02-42-175714-tune4 --versiontag tune5 --resume True --loadfile tune4-CP17.pth
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-07-13-02-42-175714-tune4/ --port=6006
//switch to beta=1, +weighted_bce
//Epoch: 17, Fold: 0
TrainLoss: 14.8867569101, TrainF1:
ValidLoss: 0.457575827837, ValidF1: 0.210946713931
//don't use beta=2, it is evil
=
python train.py --projecttag 2018-11-07-13-02-42-175714-tune4 --versiontag tune6 --resume True --loadfile tune5-CP23.pth
//try only with f1 loss
//focal up a lot, weighted bce down a lot, bce up little, f1 down little (from start)
//nothing happened (from CP23)
=
python train.py --projecttag tune5 --versiontag tune6 --resume False --loadfile tune5-CP18.pth --loaddir 2018-11-07-13-02-42-175714-tune4
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-08-03-24-12-983709-tune5/ --port=6006
python train.py --projecttag tune5 --loaddir 2018-11-08-03-24-12-983709-tune5 --versiontag tune7 --resume True --loadfile tune6-CP2.pth
//only focal now, add precision recall graph
python train.py --projecttag tune5 --loaddir 2018-11-08-03-24-12-983709-tune5 --versiontag tune8 --resume True --loadfile tune7-CP5.pth
//now add f1 (focall loss is bad when you just init the train)
// Epoch: 22, Fold: 0
TrainLoss: 13.4163611972, TrainF1: 0.999997869304
ValidLoss: 0.449503481388, ValidF1: 0.226643079329 Thres:0.1837, 0.6439F1
F1 by sklearn = 0.196469649036
python predict.py --loaddir 2018-11-08-03-24-12-983709-tune5 --versiontag f1andsomefocal --loadfile tune8-CP22.pth
RedstoneTorch/model/2018-11-08-03-24-12-983709-tune5/tune8-CP22.pth-f1andsomefocal-0.csv
=
python train.py --projecttag tune6 --versiontag one --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-08-13-18-39-289369-tune6/ --port=6006
python predict.py --loaddir 2018-11-08-13-18-39-289369-tune6 --versiontag bcef1-1 --loadfile one-CP18.pth
RedstoneTorch/model/2018-11-08-13-18-39-289369-tune6/one-CP18.pth-bcef1-1-F0-T0.01602.csv
// train on bce and f1
python train.py --projecttag tune6 --versiontag two --resume True --loadfile one-CP25.pth --loaddir 2018-11-08-13-18-39-289369-tune6
// adjust down lr by a factor of 10, adjust batch size by 2(32->64)
// focal loss seem to need a bigger batch size, I will see how the loss fluctuate to decide whether to add focal or not
python predict.py --loaddir 2018-11-08-13-18-39-289369-tune6 --versiontag bcef1-2 --loadfile two-CP44.pth
//python train.py --projecttag tune6 --versiontag three --resume True --loadfile two-CP46.pth --loaddir 2018-11-08-13-18-39-289369-tune6
//train +with focal
=
python train.py --projecttag normal3 --versiontag three --resume False
//add stratify fold, change focal gamma to 4, combination of loss on different stage
else
loss = f1 + bce.sum() + focal.sum()
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-11-08-00-19-978745-normal3/ --port=6006
python train.py --projecttag normal4 --versiontag one --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-13-04-54-14-153732-normal4/ --port=6006
//normalize by all data instead of just train or val
python train.py --projecttag normal4 --versiontag one --resume False --loadfile two-CP46.pth --loaddir 2018-11-08-13-18-39-289369-tune6
=
python train.py --projecttag normal4 --versiontag one --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-13-14-20-22-604828-normal4/ --port=6006
=
python train.py --projecttag normal4 --versiontag two --resume False --loadfile one-CP8.pth --loaddir 2018-11-13-14-20-22-604828-normal4
=
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-13-18-29-22-424660-normal4/ --port=6006
python predict.py --loaddir 2018-11-13-18-29-22-424660-normal4 --versiontag 0.42 --loadfile two-CP52.pth
python train.py --projecttag test --versiontag test --resume False
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-14-18-47-41-277052-test/ --port=6006
python train.py --projecttag normal4 --versiontag four --resume True --loadfile two-CP52.pth --loaddir 2018-11-13-18-29-22-424660-normal4
//fix predict path, fix prediction
python predict.py --loaddir 2018-11-13-18-29-22-424660-normal4 --versiontag 0.42 --loadfile two-CP52.pth
download: RedstoneTorch/model/2018-11-13-18-29-22-424660-normal4/two-CP52.pth-0.42-F0-T0.01.csv
download: RedstoneTorch/model/2018-11-13-18-29-22-424660-normal4/two-CP52.pth-0.42-F0-T0.5.csv
=
//fix sigmoid
python train.py --projecttag normal4 --versiontag three --resume True --loadfile two-CP52.pth --loaddir 2018-11-13-18-29-22-424660-normal4
python train.py --projecttag normal4 --versiontag four --resume True --loadfile three-CP56.pth --loaddir 2018-11-13-18-29-22-424660-normal4
//fix display, predict, add f1-gamma5, remove bce
python predict.py --loaddir 2018-11-13-18-29-22-424660-normal4 --versiontag 0.1942 --loadfile four-CP64.pth (raw threshold)
download: RedstoneTorch/model/2018-11-13-18-29-22-424660-normal4/four-CP64.pth-0.1942-F0-T0.1942.csv
python predict.py --loaddir 2018-11-13-18-29-22-424660-normal4 --versiontag 0.2187 --loadfile four-CP64.pth (smothed threshold)
download: RedstoneTorch/model/2018-11-13-18-29-22-424660-normal4/four-CP64.pth-0.2187-F0-T0.2187.csv
python predict.py --loaddir 2018-11-13-18-29-22-424660-normal4 --versiontag 0.2187 --loadfile three-CP55.pth (CP55)
download: RedstoneTorch/model/2018-11-13-18-29-22-424660-normal4/three-CP55.pth-0.2187-F0-T0.2187.csv
=
python train.py --projecttag normal5 --versiontag one --resume False --loadfile two-CP52.pth --loaddir 2018-11-13-18-29-22-424660-normal4
//change augmentation, change optimzer, output more validatuon loss, image to size 512, batch to 32
python train.py --projecttag aug --versiontag one --resume False --loadfile two-CP35.pth --loaddir 2018-11-08-13-18-39-289369-tune6
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-19-04-02-48-790862-aug/ --port=6006
//add four times TTA, adjust weighted BCE to negatively weighted, use weighted BCE, create an LB versioin submission - dropping rare class,
python train.py --projecttag test --versiontag test --resume False
python train.py --projecttag aug2 --versiontag one --resume False --loadfile two-CP35.pth --loaddir 2018-11-08-13-18-39-289369-tune6
//start training aug with 3 fold -1,2,3
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-19-18-55-35-204127-aug2/ --port=6006
=
python train.py --projecttag augnew --versiontag one --resume False
python train.py --projecttag test --versiontag one --resume False
//2 fold
=
python train.py --projecttag seresnext-augnew-2pooling --versiontag one --resume False
python train.py --projecttag seresnext-augnew-2pooling --versiontag 9ff937c --resume True --loadfile one-CP23_F[1]_PT2018-11-22-05-08-27-139778-seresnext-augnew-2pooling_VTone_LR0.1_BS64_IMG224.pth --loaddir 2018-11-22-05-08-27-139778-seresnext-augnew-2pooling
python predict.py --loaddir 2018-11-22-05-08-27-139778-seresnext-augnew-2pooling --versiontag d418e9d_THRES0.1_SK0.4862 --loadfile 9ff937c-CP27_F[1]_PTseresnext-augnew-2pooling_VT9ff937c_LR0.1_BS64_IMG224.pth
python train.py --projecttag seresnext-augnew-2pooling --versiontag 20f4aed --resume True --loadfile 9ff937c-CP27_F[1]_PTseresnext-augnew-2pooling_VT9ff937c_LR0.1_BS64_IMG224.pth --loaddir 2018-11-22-05-08-27-139778-seresnext-augnew-2pooling
RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/9ff937c-CP27_F[1]_PTseresnext-augnew-2pooling_VT9ff937c_LR0.1_BS64_IMG224.pth-d418e9d_THRES0.1_SK0.4862-F0-T0.1-LB.csv
RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/9ff937c-CP27_F[1]_PTseresnext-augnew-2pooling_VT9ff937c_LR0.1_BS64_IMG224.pth-d418e9d_THRES0.1_SK0.4862-F0-T0.1.csv
python predict.py --versiontag 4edd2fd_THRES0.268_SK0.5436 --loaddir 2018-11-22-05-08-27-139778-seresnext-augnew-2pooling --loadfile 20f4aed-CP38_F[1]_PTseresnext-augnew-2pooling_VT20f4aed_LR0.1_BS64_IMG224.pth
Download: RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/20f4aed-CP38_F[1]_PTseresnext-augnew-2pooling_VT20f4aed_LR0.1_BS64_IMG224.pth-4edd2fd_THRES0.268_SK0.5436-F0-T0.268.csv
Download: RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/20f4aed-CP38_F[1]_PTseresnext-augnew-2pooling_VT20f4aed_LR0.1_BS64_IMG224.pth-4edd2fd_THRES0.268_SK0.5436-F0-T0.1.csv
Download: RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/20f4aed-CP38_F[1]_PTseresnext-augnew-2pooling_VT20f4aed_LR0.1_BS64_IMG224.pth-4edd2fd_THRES0.268_SK0.5436-F0-T0.01.csv
python train.py --projecttag seresnext-augnew-2pooling --versiontag 4ebebfe --resume True --loadfile 20f4aed-CP38_F[1]_PTseresnext-augnew-2pooling_VT20f4aed_LR0.1_BS64_IMG224.pth --loaddir 2018-11-22-05-08-27-139778-seresnext-augnew-2pooling
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/ --port=6006
python predict.py --versiontag gsfv1 --loaddir 2018-11-22-05-08-27-139778-seresnext-augnew-2pooling --loadfile 4ebebfe-CP44_F[1]_PTseresnext-augnew-2pooling_VT4ebebfe_LR0.1_BS64_IMG224.pth
Download: RedstoneTorch/model/2018-11-22-05-08-27-139778-seresnext-augnew-2pooling/4ebebfe-CP44_F[1]_PTseresnext-augnew-2pooling_VT4ebebfe_LR0.1_BS64_IMG224.pth-gsfv1-F0-T0.268.csv
The folowing instructions are made so that you can use this library
The dataset is provided by Kaggle
However, kaggle api is not very easy to use on remote server
Please use this chrome plugging to get cookie.txt file: Here
After you upload your cookie.txt file to your remote server, use command(provided by CarlosSouza)
Please run the following command in your ~/RedstoneTorch directory
cd ~/RedstoneTorch
wget -x -c --load-cookies cookies.txt -P data -nH --cut-dirs=5 https://www.kaggle.com/c/DATASET/train.bson
wget -x -c --load-cookies cookies.txt -P data -nH --cut-dirs=5 https://www.kaggle.com/c/DATASET/download-all
The DATASET can be replaced with human-protein-atlas-image-classification
The command above will create a file named data and put your file download-all in it.
So you need to unzip the doanload-all
To do so, run the following command
unzip ~/RedstoneTorch/data/download-all -d ~/RedstoneTorch/data
and then you need to unzip the train.zip and test.zip
unzip ~/RedstoneTorch/data/train.zip -d ~/RedstoneTorch/data/train
unzip ~/RedstoneTorch/data/test.zip -d ~/RedstoneTorch/data/test
Please use sudo in front of these command if the terminal says that you don't have permissions to do so
However, you may not have the full permission to read doanloaded file, use
sudo chmod -R a+rwx train.csv
to give yourself permission to read.
By using this command
python preprocess.py
You can preprocess the data.
- You can calculate the mean and standard deviation of train and test data
- The image will transformed to .npy so that it load faster
You can start trainning by type command python train.py
Make sure you have everything setup
You can also use the following flags to train
| Flag | Function | Default |
|---|---|---|
| --projecttag | specify the project's tag | "" |
| --versiontag | specify the version's tag | "" |
| --loadfile | file name you want to load | None |
| --resume | resume or not | False |
We strongly recommend you use some tags to make sure the program runs correctly
cd ~/RedstoneTorch
python train.py --projecttag mem --versiontag mem1 --resume False
If you want to load from previous model to continue trainning progress:
python train.py --projecttag 2018-10-30-04-07-40-043900-test --versiontag test2 --resume True --loadfile test1-CP1.pth
The above information can be obtained in the command line during trainning, like this:
Validation Dice Coeff: 0.0754207968712
Checkpoint: 1 epoch; 13.0-13.0 step; dir: model/2018-10-30-04-07-40-043900-test/test1-CP1.pth
(The epoch starts from #1, whereas fold start from #0. Only Epoch got saved.)
The program use tensorboardX to display tensors
Use command
python .local/lib/python2.7/site-packages/tensorboard/main.py --logdir=~/RedstoneTorch/model/PROJECTTAG --port=6006
to open tensorboad's display on port 6006 of your server after you run train.py where PROJECTTAG can be replaced with your project tag.
Use predict.py to get the submit data table
python predict.py --projecttag 2018-10-30-04-07-40-043900-test --versiontag test2 --loadfile test1-CP1.pth
After the prediction, you probably want to download the .csv file, the directory is here:
RedstoneTorch/model/2018-10-30-04-07-40-043900-test/test1-CP1.pth-test-0.csv
# To install the Stackdriver monitoring agent:
$ curl -sSO https://dl.google.com/cloudagents/install-monitoring-agent.sh
$ sudo bash install-monitoring-agent.sh
# To install the Stackdriver logging agent:
$ curl -sSO https://dl.google.com/cloudagents/install-logging-agent.sh
$ sudo bash install-logging-agent.sh
This package depends on
matplotlib
pydensecrf
numpy
Pillow
torch
torchvision
augmentor
tensorboardX
psutil
tensorboard
tensorflow
Please use pip install to install these dependencies.
.
├── config.py
├── data
│ ├── sample_submission.csv
│ ├── test
│ │ └── [A LOT OF PICTURES]
│ ├── trian.csv
│ └── train
│ │ └── [A LOT OF PICTURES]
├── dataset
│ ├── hpa_dataset.py
│ ├── __init__.py
│ └── tgs_dataset.py
├── loss
│ ├── dice.py
│ ├── focal.py
│ ├── __init__.py
│ ├── iou.py
│ └── loss.py
├── model
├── net
│ ├── block.py
│ ├── __init__.py
│ ├── proteinet
│ │ ├── __init__.py
│ │ ├── proteinet_model.py
│ │ └── proteinet_parts.py
│ ├── resnet
│ │ ├── __init__.py
│ │ ├── resnet_extractor.py
│ │ └── resnet_model.py
│ ├── resunet
│ │ ├── __init__.py
│ │ ├── resunet_model.py
│ │ └── resunet_parts.py
│ ├── seinception
│ │ ├── __init__.py
│ │ ├── seinception_model.py
│ │ └── seinception_parts.py
│ ├── seresnet
│ │ ├── __init__.py
│ │ ├── seresnet_model.py
│ │ └── seresnet_parts.py
│ └── unet
│ ├── __init__.py
│ ├── unet_model.py
│ └── unet_parts.py
├── optimizer
│ ├── __init__.py
│ └── sgdw.py
├── pretained_model
│ ├── bninception.py
│ ├── inceptionresnetv2.py
│ ├── inceptionv4.py
│ ├── __init__.py
│ ├── nasnet.py
│ ├── resnext_features
│ │ ├── __init__.py
│ │ ├── resnext101_32x4d_features.py
│ │ └── resnext101_64x4d_features.py
│ ├── resnext.py
│ ├── senet.py
│ ├── torchvision_models.py
│ ├── utils.py
│ ├── vggm.py
│ ├── wideresnet.py
│ └── xception.py
├── project
│ ├── hpa_project.py
│ ├── __init__.py
│ └── tgs_project.py
├── README.md
├── requirements.txt
├── tensorboardwriter.py
├── train.py
├── tree.txt
└── utils
├── encode.py
├── __init__.py
├── memory.py
└── postprocess.py
16 directories, 60 files
module.layer0.conv1.weight
module.layer0.bn1.weight
module.layer0.bn1.bias
module.layer1.0.conv1.weight
module.layer1.0.bn1.weight
module.layer1.0.bn1.bias
module.layer1.0.conv2.weight
module.layer1.0.bn2.weight
module.layer1.0.bn2.bias
module.layer1.0.conv3.weight
module.layer1.0.bn3.weight
module.layer1.0.bn3.bias
module.layer1.0.se_module.fc1.weight
module.layer1.0.se_module.fc1.bias
module.layer1.0.se_module.fc2.weight
module.layer1.0.se_module.fc2.bias
module.layer1.0.downsample.0.weight
module.layer1.0.downsample.1.weight
module.layer1.0.downsample.1.bias
module.layer1.1.conv1.weight
module.layer1.1.bn1.weight
module.layer1.1.bn1.bias
module.layer1.1.conv2.weight
module.layer1.1.bn2.weight
module.layer1.1.bn2.bias
module.layer1.1.conv3.weight
module.layer1.1.bn3.weight
module.layer1.1.bn3.bias
module.layer1.1.se_module.fc1.weight
module.layer1.1.se_module.fc1.bias
module.layer1.1.se_module.fc2.weight
module.layer1.1.se_module.fc2.bias
module.layer1.2.conv1.weight
module.layer1.2.bn1.weight
module.layer1.2.bn1.bias
module.layer1.2.conv2.weight
module.layer1.2.bn2.weight
module.layer1.2.bn2.bias
module.layer1.2.conv3.weight
module.layer1.2.bn3.weight
module.layer1.2.bn3.bias
module.layer1.2.se_module.fc1.weight
module.layer1.2.se_module.fc1.bias
module.layer1.2.se_module.fc2.weight
module.layer1.2.se_module.fc2.bias
module.layer2.0.conv1.weight
module.layer2.0.bn1.weight
module.layer2.0.bn1.bias
module.layer2.0.conv2.weight
module.layer2.0.bn2.weight
module.layer2.0.bn2.bias
module.layer2.0.conv3.weight
module.layer2.0.bn3.weight
module.layer2.0.bn3.bias
module.layer2.0.se_module.fc1.weight
module.layer2.0.se_module.fc1.bias
module.layer2.0.se_module.fc2.weight
module.layer2.0.se_module.fc2.bias
module.layer2.0.downsample.0.weight
module.layer2.0.downsample.1.weight
module.layer2.0.downsample.1.bias
module.layer2.1.conv1.weight
module.layer2.1.bn1.weight
module.layer2.1.bn1.bias
module.layer2.1.conv2.weight
module.layer2.1.bn2.weight
module.layer2.1.bn2.bias
module.layer2.1.conv3.weight
module.layer2.1.bn3.weight
module.layer2.1.bn3.bias
module.layer2.1.se_module.fc1.weight
module.layer2.1.se_module.fc1.bias
module.layer2.1.se_module.fc2.weight
module.layer2.1.se_module.fc2.bias
module.layer2.2.conv1.weight
module.layer2.2.bn1.weight
module.layer2.2.bn1.bias
module.layer2.2.conv2.weight
module.layer2.2.bn2.weight
module.layer2.2.bn2.bias
module.layer2.2.conv3.weight
module.layer2.2.bn3.weight
module.layer2.2.bn3.bias
module.layer2.2.se_module.fc1.weight
module.layer2.2.se_module.fc1.bias
module.layer2.2.se_module.fc2.weight
module.layer2.2.se_module.fc2.bias
module.layer2.3.conv1.weight
module.layer2.3.bn1.weight
module.layer2.3.bn1.bias
module.layer2.3.conv2.weight
module.layer2.3.bn2.weight
module.layer2.3.bn2.bias
module.layer2.3.conv3.weight
module.layer2.3.bn3.weight
module.layer2.3.bn3.bias
module.layer2.3.se_module.fc1.weight
module.layer2.3.se_module.fc1.bias
module.layer2.3.se_module.fc2.weight
module.layer2.3.se_module.fc2.bias
module.layer3.0.conv1.weight
module.layer3.0.bn1.weight
module.layer3.0.bn1.bias
module.layer3.0.conv2.weight
module.layer3.0.bn2.weight
module.layer3.0.bn2.bias
module.layer3.0.conv3.weight
module.layer3.0.bn3.weight
module.layer3.0.bn3.bias
module.layer3.0.se_module.fc1.weight
module.layer3.0.se_module.fc1.bias
module.layer3.0.se_module.fc2.weight
module.layer3.0.se_module.fc2.bias
module.layer3.0.downsample.0.weight
module.layer3.0.downsample.1.weight
module.layer3.0.downsample.1.bias
module.layer3.1.conv1.weight
module.layer3.1.bn1.weight
module.layer3.1.bn1.bias
module.layer3.1.conv2.weight
module.layer3.1.bn2.weight
module.layer3.1.bn2.bias
module.layer3.1.conv3.weight
module.layer3.1.bn3.weight
module.layer3.1.bn3.bias
module.layer3.1.se_module.fc1.weight
module.layer3.1.se_module.fc1.bias
module.layer3.1.se_module.fc2.weight
module.layer3.1.se_module.fc2.bias
module.layer3.2.conv1.weight
module.layer3.2.bn1.weight
module.layer3.2.bn1.bias
module.layer3.2.conv2.weight
module.layer3.2.bn2.weight
module.layer3.2.bn2.bias
module.layer3.2.conv3.weight
module.layer3.2.bn3.weight
module.layer3.2.bn3.bias
module.layer3.2.se_module.fc1.weight
module.layer3.2.se_module.fc1.bias
module.layer3.2.se_module.fc2.weight
module.layer3.2.se_module.fc2.bias
module.layer3.3.conv1.weight
module.layer3.3.bn1.weight
module.layer3.3.bn1.bias
module.layer3.3.conv2.weight
module.layer3.3.bn2.weight
module.layer3.3.bn2.bias
module.layer3.3.conv3.weight
module.layer3.3.bn3.weight
module.layer3.3.bn3.bias
module.layer3.3.se_module.fc1.weight
module.layer3.3.se_module.fc1.bias
module.layer3.3.se_module.fc2.weight
module.layer3.3.se_module.fc2.bias
module.layer3.4.conv1.weight
module.layer3.4.bn1.weight
module.layer3.4.bn1.bias
module.layer3.4.conv2.weight
module.layer3.4.bn2.weight
module.layer3.4.bn2.bias
module.layer3.4.conv3.weight
module.layer3.4.bn3.weight
module.layer3.4.bn3.bias
module.layer3.4.se_module.fc1.weight
module.layer3.4.se_module.fc1.bias
module.layer3.4.se_module.fc2.weight
module.layer3.4.se_module.fc2.bias
module.layer3.5.conv1.weight
module.layer3.5.bn1.weight
module.layer3.5.bn1.bias
module.layer3.5.conv2.weight
module.layer3.5.bn2.weight
module.layer3.5.bn2.bias
module.layer3.5.conv3.weight
module.layer3.5.bn3.weight
module.layer3.5.bn3.bias
module.layer3.5.se_module.fc1.weight
module.layer3.5.se_module.fc1.bias
module.layer3.5.se_module.fc2.weight
module.layer3.5.se_module.fc2.bias
module.layer3.6.conv1.weight
module.layer3.6.bn1.weight
module.layer3.6.bn1.bias
module.layer3.6.conv2.weight
module.layer3.6.bn2.weight
module.layer3.6.bn2.bias
module.layer3.6.conv3.weight
module.layer3.6.bn3.weight
module.layer3.6.bn3.bias
module.layer3.6.se_module.fc1.weight
module.layer3.6.se_module.fc1.bias
module.layer3.6.se_module.fc2.weight
module.layer3.6.se_module.fc2.bias
module.layer3.7.conv1.weight
module.layer3.7.bn1.weight
module.layer3.7.bn1.bias
module.layer3.7.conv2.weight
module.layer3.7.bn2.weight
module.layer3.7.bn2.bias
module.layer3.7.conv3.weight
module.layer3.7.bn3.weight
module.layer3.7.bn3.bias
module.layer3.7.se_module.fc1.weight
module.layer3.7.se_module.fc1.bias
module.layer3.7.se_module.fc2.weight
module.layer3.7.se_module.fc2.bias
module.layer3.8.conv1.weight
module.layer3.8.bn1.weight
module.layer3.8.bn1.bias
module.layer3.8.conv2.weight
module.layer3.8.bn2.weight
module.layer3.8.bn2.bias
module.layer3.8.conv3.weight
module.layer3.8.bn3.weight
module.layer3.8.bn3.bias
module.layer3.8.se_module.fc1.weight
module.layer3.8.se_module.fc1.bias
module.layer3.8.se_module.fc2.weight
module.layer3.8.se_module.fc2.bias
module.layer3.9.conv1.weight
module.layer3.9.bn1.weight
module.layer3.9.bn1.bias
module.layer3.9.conv2.weight
module.layer3.9.bn2.weight
module.layer3.9.bn2.bias
module.layer3.9.conv3.weight
module.layer3.9.bn3.weight
module.layer3.9.bn3.bias
module.layer3.9.se_module.fc1.weight
module.layer3.9.se_module.fc1.bias
module.layer3.9.se_module.fc2.weight
module.layer3.9.se_module.fc2.bias
module.layer3.10.conv1.weight
module.layer3.10.bn1.weight
module.layer3.10.bn1.bias
module.layer3.10.conv2.weight
module.layer3.10.bn2.weight
module.layer3.10.bn2.bias
module.layer3.10.conv3.weight
module.layer3.10.bn3.weight
module.layer3.10.bn3.bias
module.layer3.10.se_module.fc1.weight
module.layer3.10.se_module.fc1.bias
module.layer3.10.se_module.fc2.weight
module.layer3.10.se_module.fc2.bias
module.layer3.11.conv1.weight
module.layer3.11.bn1.weight
module.layer3.11.bn1.bias
module.layer3.11.conv2.weight
module.layer3.11.bn2.weight
module.layer3.11.bn2.bias
module.layer3.11.conv3.weight
module.layer3.11.bn3.weight
module.layer3.11.bn3.bias
module.layer3.11.se_module.fc1.weight
module.layer3.11.se_module.fc1.bias
module.layer3.11.se_module.fc2.weight
module.layer3.11.se_module.fc2.bias
module.layer3.12.conv1.weight
module.layer3.12.bn1.weight
module.layer3.12.bn1.bias
module.layer3.12.conv2.weight
module.layer3.12.bn2.weight
module.layer3.12.bn2.bias
module.layer3.12.conv3.weight
module.layer3.12.bn3.weight
module.layer3.12.bn3.bias
module.layer3.12.se_module.fc1.weight
module.layer3.12.se_module.fc1.bias
module.layer3.12.se_module.fc2.weight
module.layer3.12.se_module.fc2.bias
module.layer3.13.conv1.weight
module.layer3.13.bn1.weight
module.layer3.13.bn1.bias
module.layer3.13.conv2.weight
module.layer3.13.bn2.weight
module.layer3.13.bn2.bias
module.layer3.13.conv3.weight
module.layer3.13.bn3.weight
module.layer3.13.bn3.bias
module.layer3.13.se_module.fc1.weight
module.layer3.13.se_module.fc1.bias
module.layer3.13.se_module.fc2.weight
module.layer3.13.se_module.fc2.bias
module.layer3.14.conv1.weight
module.layer3.14.bn1.weight
module.layer3.14.bn1.bias
module.layer3.14.conv2.weight
module.layer3.14.bn2.weight
module.layer3.14.bn2.bias
module.layer3.14.conv3.weight
module.layer3.14.bn3.weight
module.layer3.14.bn3.bias
module.layer3.14.se_module.fc1.weight
module.layer3.14.se_module.fc1.bias
module.layer3.14.se_module.fc2.weight
module.layer3.14.se_module.fc2.bias
module.layer3.15.conv1.weight
module.layer3.15.bn1.weight
module.layer3.15.bn1.bias
module.layer3.15.conv2.weight
module.layer3.15.bn2.weight
module.layer3.15.bn2.bias
module.layer3.15.conv3.weight
module.layer3.15.bn3.weight
module.layer3.15.bn3.bias
module.layer3.15.se_module.fc1.weight
module.layer3.15.se_module.fc1.bias
module.layer3.15.se_module.fc2.weight
module.layer3.15.se_module.fc2.bias
module.layer3.16.conv1.weight
module.layer3.16.bn1.weight
module.layer3.16.bn1.bias
module.layer3.16.conv2.weight
module.layer3.16.bn2.weight
module.layer3.16.bn2.bias
module.layer3.16.conv3.weight
module.layer3.16.bn3.weight
module.layer3.16.bn3.bias
module.layer3.16.se_module.fc1.weight
module.layer3.16.se_module.fc1.bias
module.layer3.16.se_module.fc2.weight
module.layer3.16.se_module.fc2.bias
module.layer3.17.conv1.weight
module.layer3.17.bn1.weight
module.layer3.17.bn1.bias
module.layer3.17.conv2.weight
module.layer3.17.bn2.weight
module.layer3.17.bn2.bias
module.layer3.17.conv3.weight
module.layer3.17.bn3.weight
module.layer3.17.bn3.bias
module.layer3.17.se_module.fc1.weight
module.layer3.17.se_module.fc1.bias
module.layer3.17.se_module.fc2.weight
module.layer3.17.se_module.fc2.bias
module.layer3.18.conv1.weight
module.layer3.18.bn1.weight
module.layer3.18.bn1.bias
module.layer3.18.conv2.weight
module.layer3.18.bn2.weight
module.layer3.18.bn2.bias
module.layer3.18.conv3.weight
module.layer3.18.bn3.weight
module.layer3.18.bn3.bias
module.layer3.18.se_module.fc1.weight
module.layer3.18.se_module.fc1.bias
module.layer3.18.se_module.fc2.weight
module.layer3.18.se_module.fc2.bias
module.layer3.19.conv1.weight
module.layer3.19.bn1.weight
module.layer3.19.bn1.bias
module.layer3.19.conv2.weight
module.layer3.19.bn2.weight
module.layer3.19.bn2.bias
module.layer3.19.conv3.weight
module.layer3.19.bn3.weight
module.layer3.19.bn3.bias
module.layer3.19.se_module.fc1.weight
module.layer3.19.se_module.fc1.bias
module.layer3.19.se_module.fc2.weight
module.layer3.19.se_module.fc2.bias
module.layer3.20.conv1.weight
module.layer3.20.bn1.weight
module.layer3.20.bn1.bias
module.layer3.20.conv2.weight
module.layer3.20.bn2.weight
module.layer3.20.bn2.bias
module.layer3.20.conv3.weight
module.layer3.20.bn3.weight
module.layer3.20.bn3.bias
module.layer3.20.se_module.fc1.weight
module.layer3.20.se_module.fc1.bias
module.layer3.20.se_module.fc2.weight
module.layer3.20.se_module.fc2.bias
module.layer3.21.conv1.weight
module.layer3.21.bn1.weight
module.layer3.21.bn1.bias
module.layer3.21.conv2.weight
module.layer3.21.bn2.weight
module.layer3.21.bn2.bias
module.layer3.21.conv3.weight
module.layer3.21.bn3.weight
module.layer3.21.bn3.bias
module.layer3.21.se_module.fc1.weight
module.layer3.21.se_module.fc1.bias
module.layer3.21.se_module.fc2.weight
module.layer3.21.se_module.fc2.bias
module.layer3.22.conv1.weight
module.layer3.22.bn1.weight
module.layer3.22.bn1.bias
module.layer3.22.conv2.weight
module.layer3.22.bn2.weight
module.layer3.22.bn2.bias
module.layer3.22.conv3.weight
module.layer3.22.bn3.weight
module.layer3.22.bn3.bias
module.layer3.22.se_module.fc1.weight
module.layer3.22.se_module.fc1.bias
module.layer3.22.se_module.fc2.weight
module.layer3.22.se_module.fc2.bias
module.layer4.0.conv1.weight
module.layer4.0.bn1.weight
module.layer4.0.bn1.bias
module.layer4.0.conv2.weight
module.layer4.0.bn2.weight
module.layer4.0.bn2.bias
module.layer4.0.conv3.weight
module.layer4.0.bn3.weight
module.layer4.0.bn3.bias
module.layer4.0.se_module.fc1.weight
module.layer4.0.se_module.fc1.bias
module.layer4.0.se_module.fc2.weight
module.layer4.0.se_module.fc2.bias
module.layer4.0.downsample.0.weight
module.layer4.0.downsample.1.weight
module.layer4.0.downsample.1.bias
module.layer4.1.conv1.weight
module.layer4.1.bn1.weight
module.layer4.1.bn1.bias
module.layer4.1.conv2.weight
module.layer4.1.bn2.weight
module.layer4.1.bn2.bias
module.layer4.1.conv3.weight
module.layer4.1.bn3.weight
module.layer4.1.bn3.bias
module.layer4.1.se_module.fc1.weight
module.layer4.1.se_module.fc1.bias
module.layer4.1.se_module.fc2.weight
module.layer4.1.se_module.fc2.bias
module.layer4.2.conv1.weight
module.layer4.2.bn1.weight
module.layer4.2.bn1.bias
module.layer4.2.conv2.weight
module.layer4.2.bn2.weight
module.layer4.2.bn2.bias
module.layer4.2.conv3.weight
module.layer4.2.bn3.weight
module.layer4.2.bn3.bias
module.layer4.2.se_module.fc1.weight
module.layer4.2.se_module.fc1.bias
module.layer4.2.se_module.fc2.weight
module.layer4.2.se_module.fc2.bias
module.last_linear.weight
module.last_linear.bias
| Class | BestThreshold(Raw) | BestThreshold(Smoothed) |
|---|---|---|
| All | 0.2332 | 0.2196 |
| 0 | 0.07007 | 0.1547 |
| 1 | 0.9650 | 0.1571 |
| 2 | 0.8579 | 0.1798 |
| 3 | 0.1662 | 0.1931 |
| 4 | 0.7728 | 0.1324 |
| 5 | 0.01001 | 0.1926 |
| 6 | 0.01201 | 0.09215 |
| 7 | 0.0030030 | 0.1843 |
| 8 | 0.7978 | 0.1669 |
| 9 | 0.01602 | 0.09612 |
| 10 | 0.1982 | 0.1602 |
| 11 | 0.5325 | 0.1286 |
| 12 | 0.2152 | 0.1722 |
| 13 | 0.03103 | 0.1544 |
| 14 | 0.004004 | 0.04645 |
| 15 | 0.04304 | 0.06961 |
| 16 | 0.005005 | 0.1499 |
| 17 | 0.003003 | 0.06373 |
| 18 | 0.09810 | 0.1001 |
| 19 | 0.04204 | 0.1706 |
| 20 | 0.01101 | 0.1264 |
| 21 | 0.01101 | 0.1121 |
| 22 | 0.01702 | 0.08679 |
| 23 | 0.000 | 0.000 |
| 24 | 0.03504 | 0.08634 |
| 25 | 0.01502 | 0.1221 |
| 26 | 0.0050050 | 0.1943 |
| 27 | 0.01502 | 0.1180 |
| Input Image Size | Speed | Batch Size | Format | Device |
|---|---|---|---|---|
| 4x1728x1728 | 1.16s/img | 1 | jpg | 16CPU, 1 Nvidia Tesla P100 |
| 4x512x512 | 0.0128s/img | 64 | npy | 16CPU, 1 Nvidia Tesla P100 |
| 4x512x512 | 0.0769s/img | 1 | npy | 16CPU, 1 Nvidia Tesla P100 |