What is the image data volume required while transfer learning any object detection model?

I want to do transfer learning for object detection, so I have more than 5 new custom objects which is not included in coco or imagenet dataset classes, I am planning to gather each class 6000 images, for transfer
learning of like yolov3 spp model for object detection, I am in right direction? please help

If you’re fine-tuning (transfer-learning) using a pre-trained model by just modifying last 1 or few layers, as you say, a dataset with 6000 images per class would be okay. Maybe you could do augmentation to increase the dataset by 4X-5X .

Also remember that you’re saying that it’s an object detection task, so you’ll have to draw the annotation labels (bounding boxes and classes) for the images yourself manually.