tensorboard issue: acc and fps on valide set is somewhat wrong

[jkfx]

facenet-pytorch/models/utils/training.py

Line 134 in 555aa4b

writer.add_scalars(metric_name, {mode: metric})

Is there need add writer.iteration += 1 and modifiy the writer.add_scalars(metric_name, {mode: metric_batch}, writer.iteration) as the orginal code is not pass in writer.iteration?

Beacause I find that the tensorboard generating figures in valid set have no valid acc and fps curve rather a single point.

[vcadillog]

Hi, I know this is old issue, but I found the solution to this, as you said it's not taking the writter iteration right, but it's because it's not adding it to the writer, so the line :
writer.add_scalars(metric_name, {mode: metric})
Change it for:
writer.add_scalars(metric_name, {mode: metric}, writer.iteration)
Hope it's not too late to help or if the devs can make this change.

[jkfx]

Hi, I know this is old issue, but I found the solution to this, as you said it's not taking the writter iteration right, but it's because it's not adding it to the writer, so the line : writer.add_scalars(metric_name, {mode: metric}) Change it for: writer.add_scalars(metric_name, {mode: metric}, writer.iteration) Hope it's not too late to help or if the devs can make this change.

Thanks for reply! I have solved this solution ❤