Exploiting-Web-Images-for-Weakly-Supervised-Object-Detection

Bing dataset: https://drive.google.com/open?id=0B2YzfMaCv0u0WnNodjRJajgtQ2s

Folder content:

1. Folders with images

2. bing_imdb.mat:

    imdb =
   
    	imageDir: './Bing_attr_rel_dataset'
   
    	classes: [1x1 struct]
   
    	images: [1x1 struct]
   
   
   
    imdb.images =
   
    	name: {1x34183 cell}
   
    	label: [20x34183 double]
   
    	boxes: {1x34183 cell} # Edgebox proposals
   
    	boxScores: {1x34183 cell} # Edgebox proposal scores
   
    	clean_scores: [1x34183 double] # Semantic relevance scores
   
    	diff: [1x34183 double] # Edge strength
   
    	size: [34183x2 double]