search

rainmaker22 / SMART

[NeurIPS 2024] SMART: Scalable Multi-agent Real-time Motion Generation via Next-token Prediction
Apache License 2.0
76 stars 12 forks source link

np.concatenate((cur_polyline[:, 0:3], cur_polyline[:, 3:]), axis=-1) appears redundant #18

Closed hzxie closed 3 weeks ago

hzxie commented 3 weeks ago

It appears that the code cur_polyline = np.concatenate((cur_polyline[:, 0:3], cur_polyline[:, 3:]), axis=-1) may not be necessary, as it seems to be reassembling cur_polyline without making any changes.

https://github.com/rainmaker22/SMART/blob/55947594ed2e499010cfc03a5d3d45e128ab3ec4/data_preprocess.py#L535

https://github.com/rainmaker22/SMART/blob/55947594ed2e499010cfc03a5d3d45e128ab3ec4/data_preprocess.py#L547

https://github.com/rainmaker22/SMART/blob/55947594ed2e499010cfc03a5d3d45e128ab3ec4/data_preprocess.py#L558

https://github.com/rainmaker22/SMART/blob/55947594ed2e499010cfc03a5d3d45e128ab3ec4/data_preprocess.py#L579

https://github.com/rainmaker22/SMART/blob/55947594ed2e499010cfc03a5d3d45e128ab3ec4/data_preprocess.py#L588

OrangeSodahub commented 3 weeks ago

So what does that matter?

hzxie commented 3 weeks ago

Nothing affects. But makes me confused.

rainmaker22 commented 3 weeks ago

Please disregard this part. Initially, cur_polyline had many features, and we removed some unnecessary ones by indexing and concatenation. However, due to later code modifications, the indices used for concatenation changed, so this approach no longer serves a purpose.

hzxie commented 3 weeks ago

I see. Thanks for your reply.