semantic_segmentation 4833 test images?

Thanks for sharing. I'm refining my implementation of FCN-VGG16 using KITTI road dataset. But I don't find there are so many test images in the test set. There are just a couple hundred of them. I tried to download the dataset again from KITTI to see if they updated it. No luck.

Do you mind to elaborate on that? Or if you get them from other places, mind to share? Thanks.

Dec 01 '18 11:12 Xiaohong-Deng

You're right, there are 290 test images. But you can use e.g. ffmpeg to break a video into a sequence of images and test your own data set. That's what I did.

Jan 23 '19 08:01 martsalz

You're right, there are 290 test images. But you can use e.g. ffmpeg to break a video into a sequence of images and test your own data set. That's what I did.

@martsalz Thanks. BTW, How do you label your data?

Jan 23 '19 14:01 Xiaohong-Deng

That's a good question I've asked myself. Nowhere is a guide for it. This has already been discussed here:

https://stackoverflow.com/questions/47964716/how-to-format-a-data-set-for-fully-convolutional-networks

Possibly helpful: http://www.cvlibs.net/datasets/kitti/eval_road.php

In my understanding, the labeling is only necessary for the training of own pictures.

Jan 23 '19 15:01 martsalz