search

wayveai / fiery

PyTorch code for the paper "FIERY: Future Instance Segmentation in Bird's-Eye view from Surround Monocular Cameras"
https://wayve.ai/blog/fiery-future-instance-prediction-birds-eye-view
MIT License
557 stars 85 forks source link

Lyft dataset #24

Open kaanakan opened 2 years ago

kaanakan commented 2 years ago

Hi,

can you give a brief information about how to train and evaluate on the Lyft dataset? How to structure the dataset so that the dataloader can load the samples?

Thanks in advance.

anthonyhu commented 2 years ago

Hello,

The lyft dataset should be structured as:

lyft_dataset/
    trainval/
        images/
        maps/
        train_data/

Then specify in the config file the path to cfg.DATASET.DATAROOT=lyft_dataset/

kaanakan commented 2 years ago

Thanks for the fast response. Can you also help with downloading the dataset in the appropriate format? I think Lyft is stored as zarr format but the fiery dataloader needs images right?

Thanks again.

anthonyhu commented 2 years ago

At the time of the project, I downloaded the data from the Lyft perception dataset (https://level-5.global/data/) and it came formatted as described above. Has it changed since then?

kaanakan commented 2 years ago

I think it is still the same format, I first thought you are using the prediction dataset. The perception dataset contains the folders that you have mentioned. I believe you combine both train and test-val split into one folder named trainval, right?

I have another question about the indices that you are using for the Lyft dataset. The naming convention of the images is for example host-a004_cam0_1232815252251064006.jpeg, and I believe 004 represents the index of that image. When I checked the indices in the whole dataset, there are much fewer indices than you are using (lyft_splits.py). Can you clarify these points?

Thanks in advance.