search

yancie-yjr / scd.github.io

SCD: A Stacked Carton Dataset for Detection and Segmentation
Apache License 2.0
57 stars 7 forks source link

dataset split: train, validation and test? #7

Closed antoine-avek closed 1 year ago

antoine-avek commented 1 year ago

hello, First thank you for making the dataset available to public for educational purposes. I have used LSCD+OSCD 1 label with the training and validation split proposed in the zip files in this depo. Using Azure ML and yolov5 existing model, a model could be training achieving the following metrics:

Log loss Min: 1.2727, Max: 2.5639, Last: 1.2727 log_loss_train Min: 1.2509, Max: 2.6741, Last: 1.2509 Mean average precision Min: 0.89987, Max: 0.95319, Last: 0.95041 precision Min: 0.62281, Max: 0.84570, Last: 0.84570 recall Min: 0.94030, Max: 0.96451, Last: 0.96101

I was wondering if a test dataset existed, that could use to advise on the final accuracy of the model after completing the training phase.

Many thanks

Antoine

yancie-yjr commented 1 year ago

we do not build a test set, you can split one for testing.