How can i use the trained weight to test one single image and plot the result 3D boxes in the picture
and by the way, what's differences between model_pkl and optim_pkl thanks~
question 2: can you provide a detailed procedure about how can i train the network with my own custom dataset, what kind of modification should i make in the config files, thank you very much.
i am trying to train M3D-RPN with my own dataset (similar to kitti dataset), and i encounter a problem shown in the picture below:
i read the code and find out that when using the dataset prior to generate 3D anchors, if no boxes are assigned to a specific anchor, error above may show up. So i modify the code a little, when no boxes are assigned to a specific anchor, that anchor will no be used.
i can successfully run the training code right now, hopefully i will not encounter other problems and get a good training result.
[problem solved!]
Dame, it seems that if i ignore specific anchor, error will show up when i want to restore checkpoint or test the model
i use 18 anchors when generating anchors and due to no boxes are assigned to specific anchors, 8 anchors are ignored, so in the end, only 10 anchors are put into use. Training is ok but when loading trained weight, errors above will show up. Does author have any solution to my problem?
Hi, did you ever find a solution for testing and plotting results?
hello @kevin0525 @nikola310 did you find solution to test and plot the result?
Unfortunately I haven't