Only 100 predictions outputs for "TorchVision Object Detection Finetuning Tutorial""

vision
1

When I follow the tutorial “https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html” and I make the number of predictions it is only 100? is that due to the setting in pytorch/torchvision, it is not the model?
img, _ =dataset_test[1]# dataset[0]

put the model in evaluation mode

model.eval()

with torch.no_grad():
prediction = model([img.to(device)])

prediction[0][“scores”]
#this only gives a 100, which is not enough for the application I will use it for.

2

all the 100 predictions have a confidence threshold of above 0.5, so it might be a thresh hold setting?

3

That might be the case and you could try to change these threshold with these arguments.

4

Thanks a lot, this was set to 100 by default, which might make sense in most cases.
box_detections_per_img (int): maximum number of detections per image, for all classes.
Initialize the model with, say 500 in this way
model(box_detections_per_img=500)

kind regards Emil