Awesome-resources-for-better-reading-of-computer-vision-papers.md

Awesome resources for better reading of computer vision papers

Are you starting a PhD in deep-learning-based computer vision ?

Fast and efficient paper reading is critical for your PhD success

• A PhD can be a race against time (you have to stay up-to-date with recent publications, you have to match state of the art results which is getting more and more competitive as time goes, and you have to finish writing your thesis before your funding runs out!)
• Your reading has to be as efficient as can be, some tools can help you with that !

Reading resources : Keeping up with our quickly evolving field

thecvf publication search engine:For each CVPR/ECCV/ICCV conference, the searchable full list of published papers

Pros:

• reliable source of peer-reviewed computer vision papers
• likely the state-of-the-art papers you would need to compare against are among this list
• ICCV'19 papers
• CVPR'19 papers
• ECCV'18 papers
• CVPR'18 papers
• etc...

Cons:

• search results can be a bit imprecise, so it often requires some additional skimming through papers that are returned by the search

arxiv sanity: A nice wrapper over arxiv to search for recent machine learning and vision papers

Pros:

• easy query of papers with the research tool using keywords, author names, ...
• allows you to save papers in a virtual library for future reading
• good top recent functionality based on what people put into their library (which I found a more reliable metric then hype which is based on tweets

Cons:

• only covers the last three years, so only useful to find recent papers

awesome paper lists on github : Lists of "awesome" papers and resources specific to a subject

Pros:

• Exist on a large variety of topics, and at different granularitiees, don't know where to start ? Try a search on github with awesome list and some additional keywords
• Can allow you to painlessly keep-up with papers of a very specific subfield (as in the hand-pose-estimation case)
• For some examples checkout awesomep-deep-vision and awesome-computer-vision

Cons:

• No guarantees (might not be actively maintained)
• No standardized format (can be ordered according to time, popularity, ...)
• Might not exist for your specific sub-domain of interest (for instance, I couldn't find a body-pose awesome list), this is the opportunity to create one ! It will surely benefit the community.

Aleju's papers: Alexander Jung's paper summaries of a large number of popular deep learning papers. His summaries can help you understand many popular deep learning papers in a fraction of the full paper's reading time

Pros:

• Covers a lot of papers, probably any fundation paper of deep learning (batch-norm, adam, ...), a lot of widely-used papers (Mask-RCNN, Cycle-GAN, ...)
• If you want to understand some hype or widely used paper, chances you will find it there, and if you find it it will help you understand the general ideas as well as the implementation details

Cons:

• Of course, as this effort mainly relies on the effort of one (very productive) person, not all papers are available in this cheat-sheet format

arxiv vanity: A tool to read arxiv papers in your browser, useful to read papers on your smartphone !

Pros:

• Simple interface which converts arxiv papers to webpages for easy scrolling

Cons:

• Good reading, but nothing more, as there are no easy tools to annotate the paper on your smartphone directly