Ground truth velocities are left uninitialized as NaNs

[peiyunh]

I believe this is related to #6 #42 #43 and #19 .

I followed INSTALL.md and installed nuscenes from https://github.com/poodarchu/nuscenes.git. I have also run create_data.py accordingly. From what I have seen, ground truth velocities that are cached in infos_train_10sweeps_withvelo.pkl are all NaN. I believe this is at least one of the issues that results in NaN losses.

I think line 516 in nusc_common.py

velocity = np.array([b.velocity for b in ref_boxes]).reshape(-1, 3)

should be:

velocity = np.array([
    nusc.box_velocity(token) for token in sample['anns']
]).reshape((-1, 3))

Otherwise the function (box_velocity) that computes velocity will never be called and b.velocity will stay uninitialized as NaNs.

[peiyunh]

Closing this issue as b.velocity are no longer NaNs once I properly installed the customized nuscenes.

[muzi2045]

the author use the their fork version of nuscens-devkit. you can just modify this file ~/.local/lib/python3.6/site-packages/nuscenes/nuscenes.py

# Cannot estimate velocity for a single annotation.
if not has_prev and not has_next:
     return np.array([0.0, 0.0, 0.0])

if time_diff > max_time_diff:
    # If time_diff is too big, don't return an estimate.
    return np.array([0.0, 0.0, 0.0])
else:
    return pos_diff / time_diff

[peiyunh]

Interesting. Thanks for pointing this out!

[hz3014]

Closing this issue as b.velocity are no longer NaNs once I properly installed the customized nuscenes.

Hi, I have both official and customized nuscenes devkit installed. How do you make sure that the customized version is used?

[peiyunh]

I think the easiest way is to create a virtual environment with only the customized devkit installed.