Week 4 Report
Lecture review
ai competition (posts 1~6)
https://velog.io/@naem1023/series/Ai-competition
Assignment process / deliverables
Training plan: https://velog.io/@naem1023/TIL-train-%EA%B3%84%ED%9A%8D-%EC%A0%95%EB%A6%AC-2021.08.24
Assignment process notes: https://velog.io/@naem1023/TIL-%EC%BD%94%EB%94%A9-%EC%A0%95%EB%A6%AC-2021.08.2527
Peer session summary
Since there was a competition, we shared methods for training. For example, how to set hyperparameters. Whether using a particular model even makes sense.
The internal conclusion was that the choice of CNN model didn’t make much difference. Whether mobilenet, resnet, or efficientnet, there were meaningful differences below the decimal point, but nothing decisive.
We agreed that trying various approaches is better. We’d learned a lot, and we needed to actually apply what we’d learned.
Study retrospective
21/08/23: Experimented with various methods via jupyter notebook and submitted to the competition 21/08/24: Converted notebook experiments into a pipeline and structured the project 21/08/25: Tried efficientnet-b7, VOLO, BiT, CaiT 21/08/26: Simple testing with efficientnet-b4, resnet18 while improving the project 21/08/27: Implemented CutMix