why did my detection output so weak unlike fp_det.gif showed?

[ryanyej]

I coverted default onnx model to tensorrt, and using waymo newest dataset, which is "individual_files_validation_segment-10203656353524179475_7625_000_7645_000_with_camera_labels.tfrecord". my result are shown as below few or no detection box. and your result showed good detection result as below Could you please show me what cause this different result?

[HaohaoNJU]

Have you concerted the raw data as suggested in tools/generate_input_data.py ?

[ryanyej]

yes I did . I used below commands:

1. python generate_input_data.py --data_path '~/Code/CenterPoint_tensorrt/lidars/val/lidar' --save_path '~/Code/CenterPoint_tensorrt/lidars/val/points', and I get seq_0_frame_X.bin file. (X from 0 to 197).
2. Then I tried ../build/centerpoint --pfeEnginePath=pfe_fp.engine --rpnEnginePath=rpn_fp.engine --savePath=results --filePath=/lidars/val/points --loadEngine to get seq_0_frame_X.bin.txt result. (X from 0 to 197).
3. Last , rosrun waymo_track waymo_track.pyto get visualization. And I changed waymo_track.py 's ROOT_PATH to result path which contain (seq_0_frame_X.bin.txt).

I don't know what went wrong

[HaohaoNJU]

Hi, I would advise you first to feed the provided bin files in lidars/*bin into TRT engines and show the results in the point cloud, if that's ok then you will be sure that your generated bin files are defective. Please convert tfrecord files to pkl files according to this, then convert the pkl files to bin files.

[ryanyej]

OK. Could you please provide lidars/*bin lidar and anno pickle files? I notice without these 2 files, can't run waymo_track.py

[HaohaoNJU]

Sure, you can see here :

link:https://pan.baidu.com/s/1Ua9F3eFflA9Gckpa9U-1Eg  passwd:08s6

[ryanyej]

Thanks ! This project is very impressive!

[mhmmdjafarg]

Hi @ryanyej were you successfully generate bin files using generate_input_data.py? How was the result? Mine is so weak accuracy when try use waymo_eval.py

[ryanyej]

HI @mhmmdjafarg how about use below datasets, this dataset can get good result as above showed.

Sure, you can see here :

link:https://pan.baidu.com/s/1Ua9F3eFflA9Gckpa9U-1Eg  passwd:08s6

[mhmmdjafarg]

Yea it shows good detection results for the bin files, but if i try to generate bin files using generate_data.py, get different results and so bad.

I dont know why @ryanyej