Yolov4 - evaluation differences

Accelerated Computing Intelligent Video Analytics TAO Toolkit
#1

• Hardware (Xavier)
• Network Type (Yolo_v4)
• TLT Version v3.22.05-tf1.15.5-py3
• Training spec file
yolo_v4_cspdarknet19.txt (2.9 KB)

Hi!
When evaluating Yolov4 model on validation dataset, the mAP is significantly different than the score obtained during validation. Using other architectures like faster-rcnn and detectnetv2 evaluation output of other architectures seems to be consistent with validation during training. Do you know what might have caused that behaviour with Yolov4?

#2

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

It does not make sense. Could you double check? Can you share the full training log and evaluation log?

#4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.