search

NVIDIA-AI-IOT / Lidar_AI_Solution

A project demonstrating Lidar related AI solutions, including three GPU accelerated Lidar/camera DL networks (PointPillars, CenterPoint, BEVFusion) and the related libs (cuPCL, 3D SparseConvolution, YUV2RGB, cuOSD,).
Other
1.3k stars 230 forks source link

Exporting PTQ model on my custom bevfusion trained weights #139

Open sandeepnmenon opened 1 year ago

sandeepnmenon commented 1 year ago

Thank you for the amazing work. I was able to setup the BEVFusion inference using the model files given in the readme. I want to use this pipeline for BEVFusion trained on my dataset, so as per the Quantization README,

  1. Run the export scripts for each module with the checkpoint as my checkpoint that I get from BEVFusion training pipeline
  2. Run qat/ptq.py with the dataset changed to my custom dataset

TLDR Can I directly use my BEVFusion weights trained on my dataset in the quantisation flow mentioned in this repo with the change in the qat/ptq.py file to use my custom dataset

cdefg commented 1 year ago

same question, same failure Need an answer, the same.

sandeepnmenon commented 1 year ago

Seems like the qat/ptq.py is written for the resnet50 backbone. So I think even for custom dataset, as long as the model architecture doesn't change, the script should work as is (except for the change in the dataloaders)