Not evaluation with backbone DenseNet

[mariaculman18]

I trained a model using as a backbone the DenseNet-121. I am trying to evaluate the model's mean average precision (mAP) with the training dataset but I am getting mAP: 0.0000.

Versions:
RetinaNet 0.5.1
Keras 2.2.4
TensorFlow GPU 1.12

While training the model I got the training loss, the validation loss and the validation mAP by using --compute-val-loss and --val-annotations dataset/test.csv. I converted the trained model to an inference model with retinanet-convert-model --backbone=densenet121 [path to trained model (*.h5)] [path to save inference model (*.h5)]. Now I want to obtain the mAP with the training dataset, so I am doing retinanet-evaluate csv dataset/train.csv dataset/classes.csv [path to inference model (*.h5)] --backbone=densenet121. But I am getting mAP: 0.0000.

Any idea what can I do?

[hgaiser]

Hey, as you have already seen, this issue is discussed in #647 . Let's keep the discussion there.