12
5/15/17 | 1 Since 1614 Scale selection in Convolutional Neural Networks using Multi - Scale Filters and Dimensional MinMax Pooling Gereon Vienken Sander Bohte Max Losch Steven Scholte

1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17 | 1

Since 1614

Scale selection in Convolutional Neural Networks using Multi-Scale Filters and Dimensional MinMax Pooling

Gereon Vienken Sander Bohte Max Losch Steven Scholte

Page 2: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Overview

›Project description›Results›Lessons learned

Page 3: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

CNN

https://devblogs.nvidia.com/parallelforall/deep-learning-nutshell-core-concepts/

Page 4: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Pooling

›Filters

http://vaaaaaanquish.hatenablog.com/entry/2015/01/26/060622

Page 5: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

MinMax

›Select optimal filter scale

›Smallest Filter with relevant information

›Threshold for value distribution

Page 6: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

MinMax

›Original Implementation by Bohte and Gehreab›Shows Min approximates optimum›Normalized Input›Pre-implemented absolute Gabor Filters

Page 7: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Implementation

Page 8: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Results

Page 9: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Results

Page 10: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Conclusion

›Slight overfitting of Min in the end›Min performs slightly better then Max›Threshold not as important

Page 11: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Lessons learned

›Start on Dataset with smaller images› Runtime: 3Days -> 5 Hour

›Make sure you understand restrictions of ground research

› Normalised images›If you use different machines make sure you understand the influence

› Overfitting accoutred on GPU cluster and weak GPU but not on TitanX

›Be aware: Drucker’s Law!› "If one thing goes wrong, everything else will, and at the

same time."

Page 12: 1 Scale selection in Convolutional Neural Networks using ...€¦ · Scale selection in Convolutional Neural Networks using Multi-Scale Filters and ... not on TitanX ›Be aware:

5/15/17

Questions

http://www.consciousentities.com/2014/02/the-hard-problem-problem/