fshamshirdar/deepvo

Deep Monocular Visual Odometry using PyTorch (Experimental)

deepvo's Issues

KITTI dataset

hi, I am new to visual odometry and want to run this code. which one of the Kitti datasets were used in the code?

Training problem

Hello! The training set I use is 00, 02, 05, 08. I don’t know what the reason is during training. There are always more than 1 pictures not trained. Do you know what the reason is?

image_2

DeepVO/dataset.py

Line 61 in 7a8e73f

image_path = os.path.join(self.base_path, 'sequences', sequence, 'image_2', '%06d' % index + '.png')

In this line the folder referenced is image_2 but when I downloaded the KITTI dataset I only get image_0 and image_1 folder.
Can you help?

Thank you

Dumping the entire sequence into GPU causes memory out error.

In def__getitem__(self, index), dataset.py you are stacking the entire train, test set sequences and returning it back. How ever getitem iterable and returns images with labels over index.

Can you replicate the result of the original paper？

Hello,
I cannot replicate the result of the original paper. If you can do it, please upload the pretrained weights. Thanks!
Because I trained it for 100 epochs, the network almost learned nothing. The testing loss is too high and the results are not visually convincing at all. Have you found a way to solve it? Thanks!