Hello everyone, greetings!
I created an annotation file in .csv format for a multilabel classification object detection model. I am using the FasterRCNN algorithm for the training. I tried to annotate the image and labels in the .csv file in the format where columns are [image_id, class_name, class_id, x_min, y_min, x_max, y_max]. Since some of my image datasets have two objects in the image, the approach I tried is that I assigned two different rows in the .csv file and annotate the class and bbox info for each class for the same image. Here I think my model is considering the same image as two different independent images for their respective classes. (am I correct?)
After having the closer look at the validation output, I observed that the image with two objects inside only predicts one object in the image (and that is just Obj_B) whereas the ground truth says two objects (Obj_A & Obj_B) inside the image. I am unable to investigate, why the model predicts only one object in the image (i.e.just Obj_B)? have I annotated the information in .csv in an incorrect way? Could anyone provide/show me a typical/sample .csv file that has been used for multilabel classification obj. detection case? I would be also thankful for your overall opinion or suggestion in this regard.
(wish you all a good day and sry for my bad english)