search

ydhongHIT / DDRNet

The official implementation of "Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes"
MIT License
415 stars 53 forks source link

About camvid #25

Closed xiaodaomingye459 closed 2 years ago

xiaodaomingye459 commented 2 years ago

Hello, I noticed that the prediction result of your paper in the Camvid data set is marked with black ignore label. May I ask how you did it? If my prediction result only has 11 categories, there will be no such effect.

ydhongHIT commented 2 years ago

Hello, I noticed that the prediction result of your paper in the Camvid data set is marked with black ignore label. May I ask how you did it? If my prediction result only has 11 categories, there will be no such effect.

I didn't understand what effect you get. I just set the ignored pixels to black.

xiaodaomingye459 commented 2 years ago

Here is the prediction result，there are only 11 categories, what can I do to get the prediction like yours. image Maybe I don't know much about it，whether you can publish your dataset of camvid. My email is 755435459@qq.com Thank you very much indeed.

ydhongHIT commented 2 years ago

Here is the prediction result，there are only 11 categories, what can I do to get the prediction like yours. image Maybe I don't know much about it，whether you can publish your dataset of camvid. My email is 755435459@qq.com Thank you very much indeed.

The original CamVid consists of 32 classes. This is because during training, I ingored the other 21 classes instead of merging the 32 classes into 11 classes. So when getting the mIoU, I follow the same strategy. And if you want to get the visualized results with only 11 classes, you can directly visualized the network output without considering the ignored pixels. The dataset I used is provided in the readme, the same as the PP-Liteseg. I hope this can help you.