Intelligent Extra Arm
Retraining the model
Adding to the last forum discussion, in this, I'll like to cover up on the retraining of the previously created object detection model. I followed below steps before retraining the model.
- I added more images of the different electronic components like resistor, capacitor, diode etc. Also, focused on adding more than one component in an image.
- Also, I limited the bounding boxes excluding the lead terminals. Below are the before and after images. Labels are case-sensitive, so make sure to name them uniformly.
- Moved failed images in testing dataset to training dataset and then to replace the gap added even more similar testing data. Under 'Model Testing' you can click on the three vertical dots against the data which you would like to move and select 'Move to training set' option.
After the above changes, I clicked on the 'Retrain model' option. The updated dataset was fed to the model, it was then trained and tested. Below screenshot shows the final outcome.
Compared to previous, it is now able to identify two components in a single image.
But there is still room for further fine tuning and re-training. As can be seen in below classification output, capacitor is classified as resistor also.
