search

umautobots / vod-converter

Convert between visual object detection datasets
MIT License
307 stars 93 forks source link

Change content format of train.txt #1

Closed nghiattran closed 7 years ago

nghiattran commented 7 years ago

Content format of train.txt from KITTI consists of 2 things: path to image file and a path to label file of that image. Even though this piece of data may seem redundant, I think it is better to have a consistent format with KITTI.

krosaen commented 7 years ago

hey, thanks for taking the time to contribute!

However, I can't find it documented anywhere that train.txt should include two paths to the image and labels; it may be that this isn't really offically part of KITTI but that we adopted a VOC like image set format so that we could work with KITTI-like datasets with different images. We prefer to stay consistent with VOC and have already developed tooling that expects this format.

So unless KITTI has a standard for having a train.txt with paths, I'd prefer to keep things as is.

nghiattran commented 7 years ago

I think they already changed Kitti format. I downloaded left color images of object data set yesterday and the train.txt actually contains two paths to the image and labels.

krosaen commented 7 years ago

Ok thanks, I'm downloading that now and will take a look.

krosaen commented 7 years ago

@nghiattran I just looked and do not see any train.txt within data_object_image_2.zip, it's only images within 2 directories.

the link from my email is http://kitti.is.tue.mpg.de/kitti/data_object_image_2.zip are you downloading from somewhere else?

nghiattran commented 7 years ago

I am so sorry for this confusion. I used a script to download data but didn't know that it automatically generates train.txt.

krosaen commented 7 years ago

No problem at all! Thanks again for taking the time in any case.