资源分类

TSN-surveillance

2020-03-31 |

35 |

0 |

TSN-surveillance

TSN-Pytorch

Using for surveillance system

Now in experimental release, suggestions welcome.

Note: always use git clone --recursive https://github.com/yjxiong/tsn-pytorch to clone original tsn-pytorch.

Training

Dataset : Trimmed UCF Crimes - https://github.com/henniekim/action_recognition_study/wiki/Temporal-annotation-for-UCF-Crimes-dataset

Note: How I set for each model is in table.

For RGB models:

	seg5	seg10	seg15	seg20
dropout 0.4	x	o	o	o
dropout 0.6	o	o	o	o
dropout 0.8	x	o	o	o

/datahdd/workdir/hyeongeun/anaconda3/envs/test2/bin/python3.6 
/datahdd/workdir/hyeongeun/test/tsn-pytorch/main.py ucf101 RGB 
/datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_trainingset.txt 
/datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_testset.txt --arch BNInception 
--num_segments 10 --gd 20 --lr 0.0001 --lr_steps 30 60 --epochs 300 -b 32 -j 2 
--dropout 0.6 --snapshot_pref re_RGB_seg10_drop06

For flow models:

	seg5	seg10	seg15	seg20
dropout 0.6	o	o	o	o

/datahdd/workdir/hyeongeun/anaconda3/envs/test2/bin/python3.6 
/datahdd/workdir/hyeongeun/test/tsn-pytorch/main.py ucf101 Flow 
/datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_trainingset.txt 
/datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_testset.txt --arch BNInception 
--num_segments 10 --gd 20 --lr 0.0001 --lr_steps 30 60 --epochs 300 -b 32 -j 2 
--dropout 0.6 --snapshot_pref re_Flow_seg10_drop06

For Warp models:

	seg10	seg15	seg20
dropout 0.6	o	o	o

/datahdd/workdir/hyeongeun/anaconda3/envs/test2/bin/python3.6 
/datahdd/workdir/hyeongeun/test/tsn-pytorch/main.py ucf101 Flow  /datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_warp_trainingset.txt  /datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_warp_testset.txt --arch BNInception 
--num_segments 10 --gd 20 --lr 0.0001 --lr_steps 30 60 --epochs 300 -b 16 -j 2 
--dropout 0.6 --snapshot_pref re_Warp_seg10_drop06

Testing

After training, there will checkpoints saved by pytorch, for example re_RGB_seg10_drop06_checkpoint.pth. Also, there will best point saved, for example re_RGB_seg10_drop06_best.pth

Use the following command to test its performance in the standard TSN testing protocol:

for RGB models:

/datahdd/workdir/hyeongeun/anaconda3/envs/test2/bin/python3.6 
/datahdd/workdir/hyeongeun/test/tsn-pytorch/test_models.py ucf101 RGB 
/datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_testset.txt  /datahdd/Dataset/Trimmed_UCF_Crimes/TSN_result/re_RGB_seg10_epoch300/re_RGB_seg10_drop06_best.pth 
--test_segments 10 --arch BNInception --save_scores re_RGB_seg10_dp06

for Flow models:

/datahdd/workdir/hyeongeun/anaconda3/envs/test2/bin/python3.6 
/datahdd/workdir/hyeongeun/test/tsn-pytorch/test_models.py ucf101 Flow 
/datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_testset.txt  /datahdd/Dataset/Trimmed_UCF_Crimes/TSN_result/re_Flow_adam_seg10/re_flow_seg10_best.pth 
--test_segments 10 --arch BNInception --save_scores re_Flow_seg10_dp06

for Warp models:

/datahdd/workdir/hyeongeun/anaconda3/envs/test2/bin/python3.6 
/datahdd/workdir/hyeongeun/test/tsn-pytorch/test_models.py ucf101 Flow  /datahdd/Dataset/Trimmed_UCF_Crimes/trim_shooting_re_warp_testset.txt  /datahdd/Dataset/Trimmed_UCF_Crimes/TSN_result/re_Warp_seg10/re_Warp_seg10_drop06_best.pth 
--test_segments 10 --dropout 0.6 --arch BNInception --save_scores re_Warp_seg10_dp06

Testing result

for RGB models:

		seg5	seg10	seg15	seg20
dropout 0.4	top1		41.71%	36.57%	40%
.	top3		59.43%	57.14%	61.14%
dropout 0.6	top1	37.71%	43.43%	41.14%	40%
.	top3	55.43%	59.43%	62.86%	60.57%
dropout 0.8	top1		40.57%	38.29%	36.57%
.	top3		61.71%	59.43%	57.71%

for Flow models:

		seg5	seg10	seg15	seg20
dropout 0.6	top1	42.29%	39.43%	35.43%	38.86%
.	top3	63.43%	60.57%	59.43%	61.14%

for Warp models:

		seg10	seg15	seg20
dropout 0.6	top1	26.29%	30.29%	30.29%
.	top3	54.86%	55.43%	53.14%

TSN-torch env pkg list

# This file may be used to create an environment using:# $ conda create --name <env> --file <this file># platform: linux-64blas=1.0=mkl
ca-certificates=2018.03.07=0
certifi=2018.4.16=py36_0
cffi=1.11.5=py36h9745a5d_0
cuda80=1.0=0
cudatoolkit=8.0=3
cudnn=7.0.5=cuda8.0_0
freetype=2.8=hab7d2ae_1
intel-openmp=2018.0.0=8
jpeg=9b=h024ee3a_2
libedit=3.1.20170329=h6b74fdf_2
libffi=3.2.1=hd88cf55_4
libgcc-ng=7.2.0=hdf63c60_3
libgfortran-ng=7.2.0=hdf63c60_3
libpng=1.6.34=hb9fc6fc_0
libstdcxx-ng=7.2.0=hdf63c60_3
libtiff=4.0.9=he85c1e1_1
mkl=2018.0.2=1
mkl_fft=1.0.1=py36h3010b51_0
mkl_random=1.0.1=py36h629b387_0
ncurses=6.1=hf484d3e_0
numpy=1.14.3=py36hcd700cb_1
numpy-base=1.14.3=py36h9be14a7_1
olefile=0.45.1=py36_0
openssl=1.0.2o=h20670df_0
pillow=5.1.0=py36h3deb7b8_0
pip=10.0.1=py36_0
pycparser=2.18=py36hf9f622e_1
python=3.6.5=hc3d631a_2
pytorch=0.3.1=py36_cuda8.0.61_cudnn7.0.5_2
pyyaml=3.12=py36hafb9ca4_1
readline=7.0=ha6073c6_4
scikit-learn=0.19.1=py36h7aa7ec6_0
scipy=1.1.0=py36hfc37229_0
setuptools=39.1.0=py36_0
six=1.11.0=py36h372c433_1
sqlite=3.23.1=he433501_0
tk=8.6.7=hc745277_3
torchvision=0.2.0=py36_0
wheel=0.31.1=py36_0
xz=5.2.4=h14c3975_4
yaml=0.1.7=had09818_2
zlib=1.2.11=ha838bed_2

上一篇：fd3d_tsn_pt

下一篇：TSN_BehaviorRecognition_GenerateTrainImages

用户评价

全部评价

还没有评论，说两句吧！

热门资源

seetafaceJNI

项目介绍基于中科院seetaface2进行封装的JAVA...
spark-corenlp

This package wraps Stanford CoreNLP annotators ...
Keras-ResNeXt

Keras ResNeXt Implementation of ResNeXt models...
capsnet-with-caps...

CapsNet with capsule-wise convolution Project ...
shih-styletransfer

shih-styletransfer Code from Style Transfer ...

智能在线

400-630-6780
聆听.建议反馈

E-mail: support@tusaishared.com