5,337
edits
No edit summary |
|||
Line 633: | Line 633: | ||
* FPN: Feature Pyramid Network | * FPN: Feature Pyramid Network | ||
===ION: Inside | ===ION: Inside-Outside Network=== | ||
Bell ''et al.'' <ref name="bell2016ion"></ref> | Bell ''et al.'' <ref name="bell2016ion"></ref> | ||
The key idea is that we want a feature vector which | The key idea is that we want a feature vector which includes multi-scale and contextual information. | ||
'''Potential Exam Question''' | '''Potential Exam Question''' | ||
Line 641: | Line 641: | ||
We want to use features from multiple levels. The RoI is fixed. | We want to use features from multiple levels. The RoI is fixed. | ||
The resolution, number of channels, and magnitude of features can be different. | The resolution, number of channels, and magnitude of features can be different. | ||
* Do L2 normalization of the features at different layers | |||
* Concatenate features | Architecture | ||
* Rescale them. | # There are 5 conv blocks, followed by two 4-dir IRNN blocks which extract context features. | ||
* | # The whole image passes through this entire network. | ||
* For each RoI identified using object proposals: | |||
* Do L2 normalization of the features at different layers (Conv3, conv4, conv5, and context features) | |||
* Concatenate features to a single feature image. | |||
* Rescale them and do 1x1 convolution to get a <math>512 \times 7 \times 7</math> feature descriptors. | |||
* Pass through two FC layers. | |||
* Finally, one FC extracts the class via softmax and another the bounding box. | |||
===Analysis and Diagnosis=== | ===Analysis and Diagnosis=== |