search

TUI-NICR / ESANet

ESANet: Efficient RGB-D Semantic Segmentation for Indoor Scene Analysis
Other
236 stars 51 forks source link

How to Dataset Inference in cityscapes‘ image #34

Open feng899 opened 2 years ago

feng899 commented 2 years ago

I runed a command：python inference_samples.py --dataset cityscapes --ckpt_path ./trained_models/cityscapes/r34_NBt1D_half.pth --depth_scale 1 --raw_depth then changed sample_rgb.png and sample_rgb.png in the test in ./samples but this segmentation is terrible. I do not know why,how can I slove this question . thanks.Look forward to your reply！

moko3016 commented 2 years ago

It depends on the images you use. The more different your images are from the dataset the model was trained on, the worse the segmentation will be.