How can I get the actual depth value of each pixel by the 3-channel depth map? about planercnn

[art-programmer]

If you are referring the groundtruth depth map from ScanNet, you could use something like this,

planercnn/datasets/scannet_scene.py

Line 135 in e773eb3

depth = cv2.imread(depthPath, -1).astype(np.float32) / self.depthShift

As for the output visualization, it's generated here,

planercnn/visualize_utils.py

Line 244 in e773eb3

cv2.imwrite(options.test_dir + '/' + str(indexOffset) + '_depth' + suffix + prediction_suffix + '.png', drawDepthImage(depth_pred[80:560]))

The input to the drawDepthImage function is the absolute depth map.