detection_head about pf-track HOT 5 CLOSED

@miraclesheephaha I am not familiar with PETRv2 right now. From the error message you just presented, it seems you have to modify the input arguments of each type of class. For the tracking purpose, I think the input to the head needs to be modified. For example, you can take a look at how I modified the things and perhaps imitate this style if you want to use PETRv2.

• Here is my modified PETR head for tracking.
• Here is the original PETR head.

from pf-track.

OK,Thank you for your reply.
I will try it.

from pf-track.

@miraclesheephaha Also, please check our documentation out to see if its information helps. https://github.com/TRI-ML/PF-Track/blob/main/documents/detection_heads.md

from pf-track.

After that, I created new petrv2_tracking_head2 and petrv2's f1_q500_800320.py, and then I added the following line to train_pipeline_multiframe in f1_q500_800320.py
dict(type='LoadMultiViewImageFromMultiSweepsFiles', sweeps_num=1, to_float32=True, pad_empty_sweeps=True, test_mode=False, sweep_range=[3,27])
According to this question, I also added generate_sweep_pkl.py of petrv2
megvii-research/PETR#60

This problem will occur as follows
PF-Track/projects/mmdet3d_plugin/datasets/pipelines/loading.py", line 63, in __call__ img_timestamp = results['img_timestamp'] KeyError: 'img_timestamp'

Looking forward to your answer,thank you.

from pf-track.

From the error message, it seems you are not using the tracking dataset, not sure if it is related, example link. Meanwhile, please also remember to prepare the data according to my documentation. @miraclesheephaha

from pf-track.