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.

The SSD300 model consistently predicted bounding boxes centered in the middle of the images, regardless of the actual content. The bounding boxes were the same for all input images, indicating a possible issue with the model’s learning or inference process.