Reading Notes - Clothing Segments / Fashion Related Papers

Paper #1 - Implementation of Real Time Dress-up System based on Image Blending
Key Notes

• Dominant colors based segmentation method
• Selected dress is scaled and rendered to fit with the subject’s body

Implementation

• Face Detection, Torso Detection
• Segmentation of region of cloth
• Dress resizing
• Dress blending

Key Approaches

• Pixel clustering and region merging
• K-means algorithm is used to cluster all the pixels values

Dress-up Algorithm

• Segmentation and replacing pixel values of frame by the pixel values of input dress image

As of 2020, With segmentation, we can detect the upper body, lower body. We can use the target dress up algorithm to replace the new design as needed.

Paper #2 - Getting the Look: Clothing Recognition and Segmentation for Automatic Product Suggestions in Everyday Photos 
Key Notes

• Detect the clothing classes present in the query image
• Image retrieval techniques to retrieve visually similar products belonging to each class found present
• Pose estimations, with the body parts depicted as colored boxes
• Clothing segmentation with Fashionista dataset
• Normalized binary mask for each segment
• For texture we used Local binary patterns (LBPs)
• For each search, we get the k most similar products
• Approximate Gaussian Mixture (AGM) clustering algorithm

Paper #3 - Animated Image Cloth Segmentation
Key Notes

• K-means clustering of image
• Segmenting the images using Gabor filters for the textured regions 

Texture based segmentation

• Input Image, Gabor Filtering, Gaussian smoothing Clustering, Segmented Image

More Reads
BACKGROUND FOREGROUND SEGMENTATION METHODS IN ANALYSIS OF LIVE SPORT VIDEO RECORDINGS

Paper - Image Based Virtual Try-on Network from Unpaired Data
Key Notes

• Synthesize images of multiple garments
• online shopping does not enable physical try-on

Novelty

• Inexpensive data collection and training process
• Online optimization capability

Steps

• Image-to-image translation network called pix2pix, that maps images from one domain to another (mouth open/closed, beard/no beard, glasses/no glasses, gender)
• Shape context
• Determine how to warp a garment image to fit the geometry 
• Convolutional geometric matcher

Networks

• Shape Generation
• Appearance Generation
• Optimization

Models

• Pose Estimation
• Segmentation
• GAN is used to warp the reference garment onto the query person image (SwapNet) swaps entire outfits between two query images using GAN
• DensePose network

Keep Thinking!!!