1. Roles of local and global information in image classification Devi Parikh

2. Coast HighwayMountain Street Forest Inside cityOpen country Tall buildings BathroomBedroomDining roomGym KitchenLiving roomMovie theater Stair case Aeroplane Car-rearFace Ketch Motorbike Watch OSR ISR CAL 2

3. Jumbled Images 3

4. Human Study: Intact Images

5. Human Study: Jumbled Images

6. Human Recognition of Jumbled Images Accuracy 6

7. How do we recognize them? Majority Vote of Local Independent Decisions? MRF Inference? … Are existing machine implementations of this model good enough? 7

8. Human Studies: Individual Blocks 8

9. Human Recognition of Individual Blocks Accuracy 9

10. Machine Experiments Assign blocks to codewords Accumulate p(scene|codewords) Pick class with highest votes

11. Majority Vote Results Proportion of matched responses 11

12. Classification of Jumbled Images Accuracy Need advanced modeling of global information 12