Colors can be specified in configs.yaml for each dataset.
vis_tools.py contains batch-wise visualization functions to modify for your own use case.
Use visualize_segmentation.py to visualize segmentation results, by providing the image with --image-path and mask (not the colored ones) with --mask-path, also --dataset needs to be specified for color selection. For detailed instructions, run:
python visualize_segmentation.py --help
For example, visualize on PASCAL VOC 2012:
python visualize_segmentation.py --image-path=test_images/voc_test_image.jpg --mask-path=test_images/voc_test_mask.png --save-path=test_images/voc_test.png --dataset=voc
You should be able to see the result like this stored at --save-path:
If mask is not provided, an inference will be performed by the model specified with --model and --continue-from, you can define input resolution with --height and --width, but the result will always be resized to the original image:
python visualize_segmentation.py --image-path=test_images/voc_test_image.jpg --save-path=test_images/voc_pred.png --model=deeplabv2 --dataset=voc --mixed-precision --continue-from=deeplabv2_pascalvoc_321x321_20201108.pt --height=505 --width=505
Use visualize_lane.py to visualize lane detection results. For detailed instructions, run:
python visualize_lane.py --help
By providing a mask with --mask-path, lanes will be drawn as non-transparent segmentation masks:
python visualize_lane.py --image-path=test_images/culane_test_image.jpg --mask-path=test_images/culane_test_mask.png --save-path=test_images/culane_test.png --dataset=culane
The result will be like this:
You can also draw sample points with --keypoint-path in the CULane format, for example:
python visualize_lane.py --image-path=test_images/culane_test_image.jpg --keypoint-path=test_images/culane_test_keypoint.txt --save-path=test_images/culane_test.png --dataset=culane
Sample points and segmentation mask can be drawn together if both files are provided.