Plotting Mean Average precision

I would like to generate plots based off the validation metrics I get every 10 epochs. Any information how I can do that will be very helpful.

Actually in ssd/dssd/yolo/retinanet jupyter notebook, you can see below cell.
It saves AP/mAP info into a csv file.
If you are running with detectnet_v2 or faster-rcnn network, you can see similar log during the training.

Now check the evaluation stats in the csv file and pick the model with highest eval accuracy.

Note csv epoch number is 1 less than model file epoch. For example, epoch 79 in csv corresponds to _080.tlt

!cat $USER_EXPERIMENT_DIR/experiment_dir_retrain/ssd_training_log_resnet18.csv
%set_env EPOCH=100

I am currently using detectnet and there is no .csv file being saved when training, there is only a monitor.json file which has the loss score per epoch, but gets updated once the epoch is completed.

Hi ishan,
During training, if you set

evaluation_config {
validation_period_during_training: 10
first_validation_epoch: 1

The mAP info will be posted at 1st epoch, 11th epoch, 21th epoch, etc.

I have that configured already, but the information is only being printed out inside the Jupyter notebook, I would like it to be stored in a csv file or some other format where I can take that data and plot charts respectively.

You can save the notebook as other format file.
For example, you click “File” → Download as “html” , then you can check all the log.

I am interested in creating plots, it is currently less intuitive to read the logs.

From the log, there is mAP info , correct?
Then , you can create the plots based on the mAP info.