0 loss when using ssd_pascal for kitti dataset

[dishita26]

Getting loss of zero while training kitti dataset on ssd

[Maharshi-Aupera]

Have you solved the problem? If yes then could you please share the solution here. I am also running into this error on my custom data-set.

[Maharshi-Aupera]

For me, the issue was the floating point conversion for the xml annotations. SSD does not read floating point numbers as the bounding box output. I was using a script to convert my ground truth annotation from .txt to .xml. But it was writing the annotation in floating point numbers. I solved that by converting the annotation values in the ground truth.