SfMLearner
SfMLearner copied to clipboard
tinghuiz
An unsupervised learning framework for depth and ego-motion estimation from monocular videos
excuse me ,can you teach me how to visualize the training process ? according to your method ,in the train.py ,I add some codes like this :grinning: ## train.py ```python...
Hi again, thanks for the code. May I ask if it is possible to test for both depth and pose simultaneously? Cheers
The pose weight --ckpt_file models/model-100280 is trained on KITTI odometry , could I directly use it to predict the pose on the other dataset such as KITTI tracking data???
Hi, Depth ground truth for testing in this code are generated from some information and parameters from the kitti raw data, while some other papers used the kitti depth annotations...
When i run eval_depth.py, here the command said need a calib_cam_to_cam.txt ,i want to know which can i find this file in the kitti website?
Hi, Since the depth net and pose net are trained jointly, they must benefit from each other. Is it possible to only train the DispNet alone as a supervised problem...
Hello tinghuiz Thanks for the code. I have read your paper,so here are some questions. In your paper,the pose net and the depth net are jointly trained. Does it means...
I have one question: 1:when i run test_kitti_pose.py ,the sequence_length=3,but the length is 5 in in pose_data of your results,why? 2:does the sequence_length matter a lot? 3 or 5 or...
Fix for missing flags, for example: "absl.flags._exceptions.UnrecognizedFlagError: Unknown command line flag 'num_source'"
Why the target image (tgt_image) and source image (src_image_stack) need to be preprocessed before depth and pose prediction? ` tgt_image = self.preprocess_image(tgt_image)` ` src_image_stack = self.preprocess_image(src_image_stack)`
