Food/Non-food Image Classification and Food Categorization using - - PowerPoint PPT Presentation

▶
food non food image classification and food
SMART_READER_LITE
LIVE PREVIEW

Food/Non-food Image Classification and Food Categorization using - - PowerPoint PPT Presentation

Jul 09, 2023 154 likes •314 views

1 Food/Non-food Image Classification and Food Categorization using Pre-Trained GoogLeNet Model Ashutosh Singla, Lin Yuan , and Touradj Ebrahimi lin.yuan@epfl.ch Multimedia Signal Processing Group Wearable, October 13, 2016 EPFL, Lausanne,

slide-1
SLIDE 1

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

1

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Food/Non-food Image Classification and Food Categorization using Pre-Trained GoogLeNet Model

Ashutosh Singla, Lin Yuan, and Touradj Ebrahimi lin.yuan@epfl.ch

slide-2
SLIDE 2

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

2

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Outline

  • Introduction
  • Image Dataset
  • Experiments and Analysis
  • Demonstration
  • Conclusion
slide-3
SLIDE 3

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

3

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Introduction

  • Dietary assessment based on multimedia

techniques, e.g., image analysis

  • Initial and crucial steps:

– Detect food images from daily images – Identify food item in a food image

  • Food categorization

– Recognizing food in major

categories may help in approx. estimation of nutritional value

slide-4
SLIDE 4

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

4

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Introduction

  • Objectives of the work

– Food/non-food image classification – Food categorization (pre-defined 11 classes.)

  • Convolutional Neural Networks (CNN) and

pre-trained GoogLeNet model

slide-5
SLIDE 5

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

5

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Dataset

  • Food-5K

– 2.5K food and

2.5K non-food

– Image source:

§ Wearable camera § Mobile phone § Existing datasets:

– Food-101 – UEC-FOOD-100 – UEC-FOOD-256

– High variety

slide-6
SLIDE 6

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

6

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Dataset

  • Food-11

– 11 major food categories – Image source:

§ Social media, e.g., Instagram, Flickr § Existing datasets:

– Food-101

– Multiple types of food in each category

slide-7
SLIDE 7

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

7

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Dataset

  • Food-11
slide-8
SLIDE 8

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

8

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Dataset

  • Food-11
slide-9
SLIDE 9

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

9

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Experiments

  • Food/Non-food Classification

– Fine tuning on the last 6 layers of a

pre-trained GoogLeNet model, on Food-5K

– 3K training, 1K validation and 1K testing – Max. acc. of 99.2%

Food Non-food Predicted classes Food Non-food Actual classes

99.4% 0.6% 1.0% 99.0%

slide-10
SLIDE 10

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

10

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Experiments

Bread Dairy products Dessert Egg Fried food Meat Noodles/Pasta Rice Seafood Soup Vegetable/Fruit Predicted classes Bread Dairy products Dessert Egg Fried food Meat Noodles/Pasta Rice Seafood Soup Vegetable/Fruit Actual classes 67.7 3.8 10.9 4.6 6.5 1.9 0.3 0.0 0.3 4.1 0.0 0.0 87.2 9.5 0.7 0.7 0.7 0.0 0.7 0.0 0.7 0.0 1.6 6.0 81.4 0.8 0.8 2.0 0.4 0.0 2.4 4.6 0.0 4.8 2.4 6.9 77.3 2.4 0.3 0.0 1.5 0.6 3.6 0.3 1.7 1.7 5.2 0.7 81.9 3.1 0.0 0.7 1.4 3.5 0.0 3.7 0.2 5.3 0.9 3.0 79.6 0.0 0.2 2.1 4.9 0.0 0.0 0.7 0.0 0.0 0.7 0.0 95.9 0.0 0.7 2.0 0.0 0.0 0.0 2.1 0.0 0.0 0.0 0.0 95.8 2.1 0.0 0.0 1.7 1.3 6.9 0.7 0.0 1.0 0.0 0.3 83.8 4.3 0.0 0.2 0.6 0.4 0.2 0.0 0.0 0.0 0.2 0.2 98.0 0.2 0.0 2.2 5.2 0.4 0.4 1.3 0.9 0.4 3.0 0.4 85.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

  • Food Category Recognition

– Fine-tune on Food-11 – GoogLeNet: last 6 layers – Best results:

§ Overall Acc. 83.5% § F-measure 0.911 § Kappa 0.816

slide-11
SLIDE 11

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

11

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Experiments

  • Food Category Recognition

– Top 10 misclassified pairs – Reasons:

§ Images within different

classes have similar appearance, shape or color.

§ Images have more than

  • ne type of food items mixed.
slide-12
SLIDE 12

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

12

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Demonstration

  • NutriTake Android App
slide-13
SLIDE 13

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

13

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Conclusion

  • Two datasets

– Food-5K: 5,000 food/non-food images – Fodd-11: 11 food categories

  • Pre-trained GoogLeNet model for

– Food/non-food classification

§ Max. accuracy of 99.2%

– Food categorization

§ Max. accuracy of 83.5%

  • A. Singla, L. Yuan, and T. Ebrahimi. Food/Non-food Image Classification and Food

Categorization using Pre-Trained GoogLeNet Model. In Proceedings of the 2nd International Workshop on Multimedia Assisted Dietary Management (MADiMa '16). Link to dataset/App:

slide-14
SLIDE 14

Multimedia Signal Processing Group Ecole Polytechnique Fédérale de Lausanne

14

Wearable, October 13, 2016 EPFL, Lausanne, Switzerland

Questions