I am trying to implement ResNet-50 and SSD300 models using PyTorch framework for object detction task. I have encountered a problem where my SSD300 model focuses only on the center of the image and predicts the same bounding box for all images. I need help identifying and solving this issue.
1-I implemented the resnet50 as backbone with SSD300 model in PyTorch framework for object detction task.
2-I followed the recommended preprocessing steps for input images, including resizing and normalization.
3-I used the pre-trained models and expected them to generalize well to the input images provided
4-i used data augmentation techniques.
I expected the SSD300 model to detect objects in different parts of the image and return varying bounding boxes based on the contents of each image. Specifically, I anticipated that the model would not focus solely on the center of the image and would provide accurate predictions for objects in various locations within the images.