Max Pooling with Vision Transformers Reconciles Class and Shape in Weakly Supervised Semantic Segmentation.
Simone RossettiDamiano ZappiaMarta SanzariMarco SchaerfFiora PirriPublished in: ECCV (30) (2022)
Keyphrases
- weakly supervised
- semantic segmentation
- superpixels
- object class
- relation extraction
- object classes
- topic models
- semi supervised
- scene classification
- conditional random fields
- computer vision
- object categories
- named entities
- multiscale
- shape model
- pascal voc
- object detection
- viewpoint
- bounding box
- co occurrence
- automatic extraction
- input image
- object recognition
- image processing