Presentation is loading. Please wait.

Presentation is loading. Please wait.

Introduction Problem: Classifying attributes and actions in still images Model:  Collection of part templates  Specific scale space locations (human.

Similar presentations


Presentation on theme: "Introduction Problem: Classifying attributes and actions in still images Model:  Collection of part templates  Specific scale space locations (human."— Presentation transcript:

1

2 Introduction Problem: Classifying attributes and actions in still images Model:  Collection of part templates  Specific scale space locations (human centric)  Discriminative learning  Sparse Activation

3 Motivation TrainTestTrainTest

4 Overview Image Scoring Mining Parts & Learning Templates

5 Formulation fractional multiples of width and height Dataset: Model: Objective:

6 Model fractional multiples of width and height... Part 1 Part 2Part 3 parts d = 1000 Model

7 Model & Scoring Image Scoring Model overlap constraint sparse activation Optimization: Greedy selection of 0.33 overlap constraint

8 Model Initialization 1) randomly sample the positive training images for patch positions: 2) Initialize model parts: perfect case: worst case: 3) BoF features normalized 10 5 patches. 3) Prunning: remove unused parts

9 Learning k = 4

10 Experiments Willow 7 Human actions 27 Human Attributes (HAT) Stanford 40 Human Actions

11 Implementation Features: – VLFeat - Dense SIFT, step size: 4 pixels square patches (8 to 40 pixels) – k-means - vocabulary 1000 – explicit feature map + Bhattacharyya (Hellinger – Square root) kernel Baseline: 4 level spatial pyramid Immediate context: – expand the human bounding boxes by 50% in both width and height Full image context: – full image classifier uses 4 level SPM with an exponential 2 kernel

12 Qualitative Results

13 Willow Actions

14 Database of Human Attributes (HAT)

15 Stanford 40 Actions

16 Learned Parts - I In each row, the first image is the patch used to initialize the part and the remaining images are its top scoring patches

17 Learned Parts - II In each row, the first image is the patch used to initialize the part and the remaining images are its top scoring patches

18 Learned Parts - III In each row, the first image is the patch used to initialize the part and the remaining images are its top scoring patches


Download ppt "Introduction Problem: Classifying attributes and actions in still images Model:  Collection of part templates  Specific scale space locations (human."

Similar presentations


Ads by Google