Clustering_algorithm HYBRID doesn't work

[NilsAI]

Please provide the following information when requesting support.

• Hardware RTX4000
• Network Type Detectnet_v2
• TLT Version:sys From this container
• Training spec file: detectnet_v2 from here TAO Toolkit Computer Vision Sample Workflows | NVIDIA NGC
• How to reproduce the issue ?

I did everything inside the container, following these guidelines https://docs.nvidia.com/metropolis/TLT/tlt-user-guide/text/object_detection/detectnet_v2.html#post-processor:

1. Follow the Notebook detectnet_v2 (removing TAO from the calls),
2. Before calling the detectnet_v2 train, go to detectnet_v2_train_resnet18_kitti.txt and change the clustering from DBSCAN to HYBRID and add the fields nms_iou_threshold: 0.2 and nms_confidence_threshold: 0.7, save.
3. Run the cell with detectnet_v2 train

NotImplementedError: Invalid clustering algorithm: hybrid

[Morganh]

Please use the latest user guide.
See

NVIDIA TAO

→

TAO Toolkit - NVIDIA Docs

• DBSCAN: Density Based Spatial Clustering of Application
• NMS: Non-Maximal suppression
• HYDRID: DBSCAN + NMS

Under HYBRID clustering, DetectNet_v2 post-processing first passes the raw network outputs to the DBSCAN clustering and uses the candidate boxes per cluster from DBSCAN as input to NMS. The NMS clustering generates the final rendered boxes.

For HYBRID clustering, ensure both DBSCAN and NMS related parameters are defined in the post-processing config.

[NilsAI]

We followed that. The problem is when we use the enum HYBRID. We get that error.

NotImplementedError: Invalid clustering algorithm: hybrid

Also, the documentation has a typo: HYDRID instead of HYBRID

I tried hydrid too, it doesn’t work. Neither DBSCAN + NMS

Just DBSCAN works fine.

[Morganh]

Could you please share your training spec file? Thanks.

[NilsAI]

detectnet_v2_train_resnet18_kitti.txt (5.6 KB)

[Morganh]

Thanks for the info. I can reproduce this error.
Unfortunately there is an issue in the code. We already have the solution and make it happen in next release.
Thanks for the catching again. Sorry for the inconvenient.