Some suggestions for graduate students in computer vision 1. Discuss with your tutor and choose a suitable research direction.
2.？ Supplement basic ability. Classic machine learning methods, deep learning foundation, network structure, training tricks and so on. Basic CV methods such as classification, detection and segmentation are well known.
3. Solid foundation, thick rooted foundation. First of all, read and master all the papers related to your research direction in the three major CV summits (CVPR, ICCV, ECCV) and important journals (TPAMI, IJCV) in recent three years. Finally, the framework and performance of methods in this field are numerous.
4. At the beginning, be familiar with several open source and performance SOTA codes in this field. Modifying on the basis of them will save you a lot of effort and make it easier to reach the new SOTA.
5. The tutor is very busy, but learn to urge the boss appropriately, have good ideas and discuss them in time, and let the boss help him revise his paper and make suggestions repeatedly. If you don't push your boss, your boss is someone else's boss.
6. Write your first paper as soon as possible, even if the hope is slim. Being familiar with the whole process of writing papers, submitting papers, refuting and camera preparation earlier will make you know more about scientific research and walk faster than your peers.
7. Ideas and papers were pushed by push itself, not by accident. Catch up with DDL and catch up with DDL is an important ability of researchers.
8. Exercise and have a good sleep. Personally, I never stay up late, and I don't think about scientific research and study for half an hour before going to bed, because I have been insomnia all night. But some people are inspired by generate in the early morning, which is very efficient, so forget this ~
9.？ Don't give up every idea easily, even if you fail, find out the reasons for your failure, and don't give up because of your poor performance. Many times, the idea of not working is separated from success by a window paper.
10. Record every inspiration, even if it is vague and rough. I used to have a notebook. When I was depressed, a dozen ideas were crossed out one by one. No one is successful, and the ability to resist setbacks is also very important.
1 1. Have the opportunity to communicate with outsiders and learn the scientific research ideas and writing styles of different schools, different research groups and laboratories of large factories, which is the best of a hundred schools.
12. Exercise your drawing and writing skills. Papers depend on the face, and the acceptance rate of good-looking papers will be much higher. In writing ability, learn more about Daxie's papers.
13.？ Keep an open mind and keen perception. Not only limited to my own small field, but also follow up the latest progress in ML, NLP, RL and other fields, always thinking about whether the stones of other mountains can be used by me.
14. Hold the thigh of the laboratory tightly. Conference journals still need a lot of experience and accumulation in submission skills, paper writing, rebel tactics and so on.