The new Lyft data

[huyuening]

How long do you expect it to take to complete the conversion code for the new Lyft level 5 predictive dataset?

[QuanyiLi]

It doesn't have a high priority for now. If everything goes well, we expect to have it in late this year.

If you want to use this data soon, you can try implementing a converter on your own. And we will provide any support we can to help you make it. As the toolkit is very like nuscenes-devkit, the converter won't be very different from the nuscenes converter, and the workload should not be very large.

[neugzy]

It seems that the nuScenes converter cannot work smoothly on old lyft data. Although the old lyft data conforms to the data format of nuScenes, some properties in the json file are still different and require a lot of modifications. I'm not sure if it's an issue with the old Lyft data I downloaded, I downloaded it from the Kaggle website（ https://www.kaggle.com/competitions/3d-object-detection-for-autonomous-vehicles/data ）It is uncertain whether it has processed the original data.

[neugzy]

In detail, the data in attribute.json and category.json is different between nuscenes and old lyft. I think just writing a type mapping file can solve the problem.