This is in response to this Faster R-CNN article. I tried my own dataset and found that once I run predictions some of the predictions are blank meaning that that a prediction was not made. And also the number of bounding boxes drawn is not 100 percent accurate based on the picture. I ran only 200 epochs with a 72 accuracy. I am assuming enough learning was not done. Any suggestions here.
This might be due to the following reasons:
a. You have less amount of training data and hence the model is unable to learn the features and hence you are not getting accurate results. So, you can try to collect more labelled data.
b. You have trained the model for less number of epochs. Try to train it for more epochs that might help the model to learn more and give better predictions.