Poor detection when training detection model with detecnet_v2 detector resnet34

• Hardware (gtx1070ti)
• Network Type (Detectnet_v2)

When I use resnet34 modulo network detecnet_v2 detector, the detection is bad. I don’t know why. can you help me?
[detectnet_v2_train_resnet18_kitti.txt|attachment]
some data formats


conf file
(upload://aiD9x2qUMRLzWz8qvU6UXn02Uwj.txt) (5.2 KB)

Could you upload the files again? Thanks.

More, please set x1,y1,x2,y2 to float format.
Refer to Data Annotation Format — TAO Toolkit 3.21.11 documentation

this file
detectnet_v2_retrain_resnet18_kitti.txt (5.3 KB)

Can you share the full training log?
What is the mAP result in the end?

i have a 3 class

fire map 30%
samll_fire 90%
smoke 40%

For detectnet_v2, please note that all the training images should have the same resolution.
May I know that the resolution of your training image?
More, how about the objects size for fire/small_fire/smoke? Are they too small?

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