https://github.com/viig99/cdiscount

Solution for Kaggle Cdiscount Image Classification, Ongoing solution.
https://github.com/viig99/cdiscount

cdiscount deep-learning image-classification kaggle keras

Last synced: 3 days ago
JSON representation

Solution for Kaggle Cdiscount Image Classification, Ongoing solution.

• Host: GitHub
• URL: https://github.com/viig99/cdiscount
• Owner: viig99
• License: apache-2.0
• Created: 2017-11-10T14:11:28.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2017-12-02T14:33:08.000Z (about 7 years ago)
• Last Synced: 2024-11-11T03:33:05.549Z (2 months ago)
• Topics: cdiscount, deep-learning, image-classification, kaggle, keras
• Language: Python
• Homepage:
• Size: 30.3 KB
• Stars: 2
• Watchers: 4
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
### Cdiscount Image Classification Challenge

Cdiscount Image Classification Challange Ongoing Solution.

## Results

Right now getting top1 accuray - 0.64 accuracy & top5 accuracy - 0.81

Will provide the weights once challenge is over.

Using an SE-Denset161 model, using intitial iterations with data augmention & last few iterations without any augmentation

use adam_accumulate or sgd_accumulate for batch accumulation, code also provide in the keras issue,

https://github.com/fchollet/keras/issues/3556

Learning rate 0-3 = 0.1

			  4-7 = 0.001

			  8-10 = 0.0001