search

Loahit5101 / PointPillars-Camera-LiDAR-Fusion

Implementation of PointPillars Network with LiDAR-camera fusion for 3D Object Detection in Autonomous Driving.
30 stars 3 forks source link

Possibility of getting pretrained models for CLOCs and Pointpillars #2

Open shayari21 opened 1 year ago

shayari21 commented 1 year ago

Hello,

I also tried the implementation of CLOCS_LQS and followed all the process needed. However the results are reallly bad say around 3d AP is 4.5. I checked that you got a good AP and also have included pointpillars.

I got the following response after 10 epochs when the cls error was stagnant at around 12%.

INFO Car AP@0.70, 0.70, 0.70: bbox AP:4.2648, 4.2198, 3.2463 bev AP:4.1331, 4.0580, 3.1242 3d AP:4.0527, 3.8060, 2.9328 aos AP:4.26, 4.18, 3.21 Car AP_R40@0.70, 0.70, 0.70: bbox AP:4.4958, 4.3517, 3.2733 bev AP:4.1676, 4.1848, 3.1503 3d AP:4.0865, 3.6632, 2.6884 aos AP:4.49, 4.31, 3.24 Car AP@0.70, 0.50, 0.50: bbox AP:4.2648, 4.2198, 3.2463 bev AP:4.2682, 4.2643, 3.2814 3d AP:4.2680, 4.2537, 3.2734 aos AP:4.26, 4.18, 3.21 Car AP_R40@0.70, 0.50, 0.50: bbox AP:4.4958, 4.3517, 3.2733 bev AP:4.4994, 4.3975, 3.3087 3d AP:4.4992, 4.3866, 3.3007 aos AP:4.49, 4.31, 3.24

So i have following questions:

  1. Is it possible for you to provide checkpoints for pointpillars and CLOCS?
  2. How many epocs did you had to run the CLOCs training to achieve the accuracy?
  3. I also was curious regarding how to obtain kitti output from mmdetection framework, as mostly it doesnt give out results in that format. (i wish to switch to multi-class)

I look forward to your response. You can also reach me out at shayarib211196@gmail.com

Loahit5101 commented 1 year ago

Hi, thank you for your interest.

  1. I have added the pretrained models in CLOC_fusion/pretrained folder.
  2. 50 - Make sure you disable NMS.
  3. I modified the output format to fit CLOC's needs. You can find the changes i made in evaluate.py.
shayari21 commented 1 year ago

Hi, thanks for your response, while trying to run the eval.py in the clocs folder(there is no evaluate.py in the clocs_folder) but in the main folder which seems to be related to pointpillars. I found that it doesnt take the .pkl file as the 3d input but is expecting a .pt file. Is that a bug as the readme, is not clear for the model and i'm using the clocs_lqs repo for reference.

Moreover, pre_process_kitti.py is also missing.

Adeel-farukh3 commented 1 year ago

Hello Dear @Loahit5101 & @shayari21, I am trying to run the eval.py and train.py commands through the CLOC_fusion directory, but it is giving me the errors: No such file or directory: './log/second/faster/1.pt' while running eval.py and 'KeyError: 0' while running train.py files.
Moreover, I made another log folder, the reason is that I cannot copy the 1.pt file inside the logs because it is encrypted, and then I put a file '000001.pt from 3D' in this folder after renaming it to '1.pt', it is giving me this error now: net.eval() AttributeError: 'dict' object has no attribute 'eval' I am attaching both screenshots. eval py error eval py train py errors