Open XFeiF opened 4 years ago
εδΊ«δ½ θ§εΎζζζγζδ»·εΌηCVη±»ζη« ο½
Papers(issues) summarized are set closed,
while papers newly collected or waited for reading are opened.
For a good note (understanding) about a paper, we follow the ten-questions principle designed by Doctor Gang Hua. Here is a good post (in Chinese) about how to boost your skill of reading papers.
The answer shall address: what are the input X (e.g. a single RGB image, an image sequence, or an RGBD image), what are the Y (e.g. pose of the human in the image) what are the constraints on X and/or Y, if any.
If it is a new problem, why does it matter? A new, meaningful, and yet challenging research problem needs a keen eye to spot, and if it is big/important enough, it may draw many people to work on it. So in a sense, it is a kind of highest innovation.
If it is not an entirely new problem, why does it still matter? When you pick up a problem to work on, you need to clearly state why it is important.
Answer to this question is to address what new knowledge is advanced in the paper. A scientific hypothesis sounds like: "If we did abc in our algorithm/dnn architecture, 90% of the case we can guarantee results def." A concrete example, "In ResNet, If we do the residue block, we expect to be able to learn much deeper networks, which leads to much higher recognition accuracy."
It is important to identify the most relevant work that inspired the work in this paper. Grouping them helps with a taxonomy helps you to build a systematic view of the research problem addressed. And finally, please memorize the name of the authors and affiliations of these related works, as they will be key people who will appreciate or criticize this work.
Please summarize the key differentiation of the paper when compared with the previous related works.
Experiments design is very important. A good experiments design shall validate all claims made in the paper. Indeed, experiments should be designed around this validation.
Dataset is an important factor in scientific research. And code helps readers to reproduce the results.
Are the claims in the paper well supported by the experimental results?
Up to this point, it should be clear if the paper made one or some solid contributions, which really refers to what knowledge is advanced.
This shall summarize your understanding of the limitations of the proposed method in the paper. Addressing these limitations are naturally future research, both from the problem definition itself and/or technical improvement. Or it could be linked to some other abstract knowledge in your cognitive model and stimulate new directions to go. This final question is the creativity part.
Welcome to share your favorite computer vision papers here, and I'll read and write some notes if I think it is interesting, too :)
You can also provide your notes/comments/blogs with the paper together! π»