manila95/autolay

A novel benchmark dataset for Monocular Layout prediction

autolay's Issues

failed to download ground truth of KITTI RAW and Odometry dataset

manila95/monolayout#28

dataset labels to image projection

Hello.
Firstly I'd like to thank you for sharing your work with dataset.
But it seems that released data is not complete: there is an instruction how to get raw images for provided labels (its clear enough) but information about relations between image and annotations planes is missed.
Proposed in dataset labels are called "bird eye view" (BEV) labels, so homography between image plane and labels plane should exist.

Is there a homography matrix or any principal points (f. ex. to use cv2.findHomography) or anything else that define relations between labels and images?
Could you, please, provide transformation algorithm from labels plane to image plane?

There is a gif in readme where the annotations are projected into the images. Is there a way to reproduce this for the whole dataset?

Best regards.

Problems about the dataset

Hi, thanks a lot for sharing your interesting and great work. We downloaded the dataset at https://autolay.github.io/index.html and there are some problems.

1. We don't seem to find the label of 'Argoverse split';
2. We don’t seem to find the label of 'KITTI split', what is the relationship between it and 'KITTI raw split', and how to download the RGB image about 'KITTI split';
3. We don't seem to find the Sidewalk category of 'KITTI raw split';
4. How many categories are there in lane labels, and what are the category IDs of ego-lane and other-lane respectively;
5. We did not find ‘eval.py’ in the Github repository.

I'm looking forward for your reply

Not found eval.py

Hi, maybe forgot to upload “eval.py”, run “python train.py”

Traceback (most recent call last):
File "train.py", line 19, in
from eval import evaluate_layout
ModuleNotFoundError: No module named 'eval'

Code and dataset availability

Nice work. Congratulation on your paper acceptance at IROS conference. Any time-line when code and Autolay dataset will be released?
Thanks

release of the entire dataset

Hi, thanks a lot for giving an explicit definition of Amodal layout estimation for autonomous driving. It's an interesting and practical task. I'd like to learn more about this task. When will you release the entire dataset? It will be very helpful.

how many decoder needed

In my understanding, the layout for each class needs a specific decoder, right? So it is a binary classification problem for each class?