Extract loss for each image in the batch

[zuliani99]

I'm wondering if it is possible to extract the class confidence and location box loss for each image in the batch, so not having a unique loss metric at the end but a tensor of loss in the such form [batch_size, 2] with each column corresponding to the class confidence and location box loss?

I'm asking this since I'm willing to add an additional module up on SSD and I need the loss for each images in the batch.