Deep screen image crop and enhance

Slides:



Advertisements
Similar presentations
Original image: 512 pixels by 512 pixels. Probe is the size of 1 pixel. Picture is sampled at every pixel ( samples taken)
Advertisements

Conditional Generative Adversarial Networks
Deeply-Recursive Convolutional Network for Image Super-Resolution
Big data classification using neural network
Generative Adversarial Nets
Deep Learning for Dual-Energy X-Ray
Video Generation with GAN
Environment Generation with GANs
DeepCount Mark Lenson.
Convolutional Neural Fabrics by Shreyas Saxena, Jakob Verbeek
Deep Predictive Model for Autonomous Driving
Krishna Kumar Singh, Yong Jae Lee University of California, Davis
Textual Video Prediction Week 2
Jure Zbontar, Yann LeCun
Classification: Logistic Regression
Understanding and Predicting Image Memorability at a Large Scale
Registration of Pathological Images
Ajita Rattani and Reza Derakhshani,
Chaoyun Zhang, Xi Ouyang, and Paul Patras
Hierarchical Deep Convolutional Neural Network
Synthesis of X-ray Projections via Deep Learning
Week 6 Cecilia La Place.
Single Image Super-Resolution
CS6890 Deep Learning Weizhen Cai
Presenter: Hajar Emami
Adversarially Tuned Scene Generation
Textual Video Prediction
Enhanced-alignment Measure for Binary Foreground Map Evaluation
A New Approach to Track Multiple Vehicles With the Combination of Robust Detection and Two Classifiers Weidong Min , Mengdan Fan, Xiaoguang Guo, and Qing.
Neural Photo Editing Andrew Brock.
Bird-species Recognition Using Convolutional Neural Network
Learning to See in the Dark
Wei Liu, Chaofeng Chen and Kwan-Yee K. Wong
Albert Xue, Binbin Huang, Jianrong Wang
Counting in Dense Crowds using Deep Learning
Road Traffic Sign Recognition
GAN Applications.
Spatial Transformer Networks
Outline Background Motivation Proposed Model Experimental Results
SVM-based Deep Stacking Networks
Lip movement Synthesis from Text
Advances in Deep Audio and Audio-Visual Processing
Convolutional Neural Network
Video Imagination from a Single Image with Transformation Generation
TPGAN overview.
Abnormally Detection
Chuan Wang1, Haibin Huang1, Xiaoguang Han2, Jue Wang1
Textual Video Prediction
VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION
Object Detection Implementations
Training-based Super Resolution Enhancement using CUDA
Weeks 1 and 2 Aaron Ott.
Deep screen image crop and enhance
Multi-UAV to UAV Tracking
Deep screen image crop and enhance
End-to-End Facial Alignment and Recognition
Report 7 Brandon Silva.
Appearance Transformer (AT)
SFNet: Learning Object-aware Semantic Correspondence
GIF2Video: Color Dequantization and Temporal Interpolation of GIF images Yang Wang, Haibin Huang, Chuan Wang, Tong He, Jue Wang, Minh Hoai. Stony Brook.
Deep screen image crop and enhance
Eliminating Background-Bias for Robust Person Re-identification
Recent Developments on Super-Resolution
Learning to Navigate for Fine-grained Classification
CRCV REU 2019 Aaron Honculada.
SDSEN: Self-Refining Deep Symmetry Enhanced Network
Deep screen image crop and enhance
Deep screen image crop and enhance
Directional Occlusion with Neural Network
Shengcong Chen, Changxing Ding, Minfeng Liu 2018
Presentation transcript:

Deep screen image crop and enhance Week 1 (Aaron Ott, Amir Mazaheri)

Problem Image Detector/Cropper Image Enhancer We have taken a photo of an image, and we want the original image. This can be broken into 2 parts: Image Detector/Cropper Image Enhancer and

Spatial Transformer Networks Jaderberg, Simonyan, Zisserman, Korayk. “Spatial Transformer Networks”, 4 Feb 2016 Localization Network – generate Theta Grid Generator – Maps pixels for transformation Sampler – Creates transformed image https://github.com/oarriaga/STN.keras

Spatial Transformer Networks (cont.) Benefits Localization network is fully trainable Module can be included in any existing model Uses: reshaping input image to better focus on important regions attention mechanism cropping images? https://github.com/oarriaga/STN.keras

Dataset 10 Images (100 Samples, 90 training – 10 validation) - (Caltech-UCSD Birds 200)

Loss Function

Results Validation Loss: 0.1195 => 0.05349 over 200 epochs (less than 5 minutes of training)

Output Input Output Ground Truth

Image Enhancement: Existing Research Photo Enhancement Super Resolution SRGAN EDSR WDSR https://github.com/krasserm/super-resolution.git Chen, Wang, Kao, Chuang. “Deep Photo Enhancer: Unpaired Learning for Image Enhancement from Photographs with GANs”, CVPR 2018 Ledig, Theis, Huszar, Caballero, Cunningham, Acosta, Aitken, Tejani, Totz, Wang, Shi. “Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network”. 25 May 2017 Lim, Son, Kim, Nah, Lee. “Enhanced Deep Residual Networks for Single Image Super-Resolution”. 10 July 2017 Yu, Fan, Yang, Ju, Wang, Wang, Huang. “Wide Activation for Efficient and Accurate Image Super-Resolution”. 21 December 2018

Next Steps Fine-tune the photo cropper Test different localization network hyperparameters Loss function and optimizer Build a more difficult data set to test the cropper on Validate on new images Further distance away More background noise Change location within image Begin working on Image Enhancement Connect Cropper to existing image enhancement network Fine tune image enhancement networks