where is 100 from?

[xxxxjj]

Hello,how does 100 in lines 52 and 53 in the model.py file come from?

[LourencoVazPato]

100 is the sequence length of the inputs. Note that the input is a sequence of 100 bounding boxes

[xxxxjj]

How is the sequence length of this input set? Why is the sequence length corresponding to each round of batch_size different when I train my data set? But the coco data set can keep the length of each round of sequence 100.

[LourencoVazPato]

The models I used to get the predictions outputed a list of 100 object predictions per image by default, are you using a different type of model?

[oayhan1]

I'm trying to train the model with a different dataset and number of object detections per image might be lower than 100. When reread the paper, I encountered a sentence like this "Sequences are padded to length 100 (the maximum number of detections often outputted by a detector). Could you explain the padding procedure? Do you add zeros to represent remaining detected bboxes to equalize the length of the sequence to 100?

[oayhan1]

I saw line to pad sequence in dataset.py. You're adding 0 to inputs and -1 to targets.