**
Goal:
**
propose an adaptive scenario discovery framework robust to various scenarios under camera perspective and physical barriers. The system is able to represent highly variable crowd images.
**
Contribution:
**
From the perspective of scenario discovery, a novel adaptive framework for crowd counting is proposed. Different from previous multiple columns/branches frameworks, ours has the ability to represent highly variable crowd images with two branches by incorporating the discretized pathway-wise responses.
**
Architecture:
**
There are two initial destiny maps separately estimated by the dense branch and sparse branch. Then the third branch named adaptive scenario discovery framework will give a dynamic weighting strategy to the two maps. After that the final density map coms out.
**
Adaptive Scenario Discovery:
**
The outputs of the last convolutional layer in the backbone go through a global average pooling(GAP) and two fully-connected layers(FC) and then have an initial response w. We expect w to adaptively recalibrate the weight of the dense and sparse pathways, therefore we normalize it into the interval of [0,1) with the following formula: