search

HRNet / HRNet-Semantic-Segmentation

The OCR approach is rephrased as Segmentation Transformer: https://arxiv.org/abs/1909.11065. This is an official implementation of semantic segmentation for HRNet. https://arxiv.org/abs/1908.07919
Other
3.13k stars 686 forks source link

Visualization result is error #191

Open hello-program opened 3 years ago

hello-program commented 3 years ago

@sunke123 @welleast @PkuRainBow hello,I used seg_hrnet_ocr_w48 to split wall and window, when I visualized, the result is error. my dataset is custom,and I create it like cityscapes. the result as follow: 082 the ground truth is : label_viz there is no window, and all of images no window my "NUM_CLASSES" of yaml file is set 3(background is 0, wall is 1, window is 2),and I modify self.label_mapping of "cityscapes.py" to self.label_mapping = {0:0, 1:1, 2:2}, and I have tried self.label_mapping = {0:ignore_label, 1:0, 2:1},but the result is error also. please help me ,thank you

hnsywangxin commented 3 years ago

my result is error also, why?

NikhilChowdary-MCW commented 3 years ago

Hi @hello-program and @hnsywangxin ,

What's the mIoU and class IoU you acheived? (If possible share a screenshot of the terminal). Both gets displayed on the terminal screen with loss also. Class IoU will be in the brackets.

It also looks like class imbalance. The black content in the picture is more compared to Red and Red is having more pixels compared to green.