For questions related to the faster R-CNN model, which was proposed in "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks" (2015) by Shaoqing Ren et al. and published in NeurIPS. Faster R-CNN is an improved version of fast R-CNN, which, in turn, is an improved version of R-CNN.
Questions tagged [faster-r-cnn]
8 questions
7
votes
2 answers
Is there an argument against using the (reviewed) predictions of a model as ground truth to further train exactly this model?
I plan to use my predictions as ground truth to continue training my model. These predictions are of course reviewed during this process. Is there an argument against that (reinforcement of slight mistakes/overfitting etc.)?
Here my specific use…
thzu
- 73
- 4
6
votes
1 answer
How does the region proposal method work in Fast R-CNN?
I read so many articles and the Fast R-CNN paper, but I'm still confused about how the region proposal method works in Fast R-CNN.
As you can see in the image below, they say they used a proposal method, but it is not specified how it works.
What…
ozoubia
- 61
- 2
1
vote
0 answers
Mask R-CNN: how is the inference done?
According to the Mask R-CNN paper and the picture below (taken from the paper), the mask branch is computed in parallel with the bbox classification and regression branches.
However in the paper they write that inference is done differently from…
orbit
- 21
- 2
1
vote
0 answers
Why are the learned offsets of anchor boxes in the RCNN object detection models scale invariant?
In the original RCNN paper (https://arxiv.org/pdf/1311.2524.pdf) and continued in later RCNN papers such as faster RCNN (https://arxiv.org/pdf/1506.01497.pdf) the learned offsets of the anchor boxes are scale-invariant. For example the learned…
phil
- 143
- 4
1
vote
0 answers
In anchor based object detection, why don't the anchors share the same weights?
After reading about YOLO V3 and Faster R-CNN, I don't understand why the weights for the regression head aren't the same across all boxes of the same size. Given that the backbone of these systems is fully convolutional, the location of the…
FourierFlux
- 783
- 1
- 4
- 14
1
vote
1 answer
In Faster R-CNN, how can I get the predicted bounding box given the neural network's output?
The RPN loss in Faster RCNN paper is
$$
L({p_i}, {t_i}) = \frac{1}{N_{cls}} \sum_{i} L_{cls}(p_i,p_i^*) + \lambda \frac{1}{N_{reg}} \sum_i p_i^* L_{reg}(t_i, t_i^*)
$$
For regression problems, we have the following parametrization
$$t_x=\frac{x -…
user31844
- 11
- 1
0
votes
1 answer
Confusion about faster RCNN neither object nor background label
I am trying to construct a faster RCNN from scratch using KERAS. I am generating the tensor which contains whether anchor at each location corresponds to object or background or neither for training the RPN.
The output tensor for the RPN is…
Abhisek Dash
- 33
- 5
0
votes
0 answers
How to design training loop in RPN?
I have a short question. I understand the concept of RPN but one small details keeps me from implementing it. How should I design the training loop given that I have to use only a subset of anchor boxes (128 positive and 128 negative). In other…
Bogdan Duduc
- 11
- 1