Grid features extraction

[ifmaq1]

1-) I have a confusion regarding grid feature extraction.
In region features we select top k bounding boxes i-e 36 or 100 depending upon the requirements.
Now grid features extract features based upon the Image size right? does it involve bounding boxes around the object as well along with the labels of object? Can you please explain that how we do the object detection/recognition part with these grid features on coco dataset?

2-) Also, I checked the tensor of one image feature file which displays [1, 2048, 19, 29]. Does it means we have total number of features 19*29=551. Can we select some top k features in it as well?

3-) lastly, does .pth features files only contains weights? or any other information as well. I am unable to read the complete file but only can see few weights