PyTorch implementation of Deep Smoke Segementation this is a implement of Deep smoke segmentation
train data with password “w5nv”,test data: DS01 with password "bymb", DS02 with password "0s0d", DS03 with password "n8w4";
you can run the test using model with password "3d7c"
- pytorch
- opencv-python
Also, this code should be compatible with Python versions 3.6 or 3.7.
run train.py to train the model
run test.py to test your trained model.You should preprocess the test dataset before test by setting a threshold in A channel （A channel means RGBA's A channel）
you can run vis.py to put the test result(0/255png)mask on the original pic as shown at first.
the result after vis is 
you can also use trained model to test real smoke,which can be more accurate if you finetune on the real smoke picture.
if you have any question ,you can pull the Issues or email me 359684740@qq.com.