Date

Jul 19, 2024

Type

with Prof.

Contributors

Notice / Previous Week

Memos

• pix2pix 모델이 이미지의 색 뿐만 아니라 이미지의 텍스쳐 등도 변화를 일으키나요?
• 아예 새로운 이미지 생성
• metric과 person perception 괴리감이 있어 보임

• 8번째 page, image에서 text로의 변환이 시각 자체랑 같은 기회를 제공하진 않는 것 같음, 11가지 bct로의 표현도 부족해보임 → 시각장애 자체가 민감한 주제라 워딩이 중요함

• text generation from image paper가 많은데 DCE paper를 사용한 이유
• 색에 집중하다 보니….
• 관련 주제에 대한 더 좋은 저널의 paper가 필요해보임
• DCE paper가 나온 곳은 신뢰성이 떨어짐

• dominant color extraction, presentation of detection result ← 2개의 main topics
• super pixel을 이용하는 등의 방법은 이미 old fashion way ← 개선은 필요 없을 듯
• richness를 어떻게 보장할 것인지 (text, color extraction 측면에서)
• 매 프레임에 대한 rich text 제공은 유저 입장의 피로도를 높일 듯
• chatgpt에서 이용하는 temperature 이용(?)
⇒ 1, 2, 3단계 등으로 조절할 수 있도록 (3으로 갈수록 더 수사적이고 비유적으로!)

• Yeonsun
• I like the second research ideation!
• Interesting reference: Berlin and Key's 11 basic color terms
• Need feedback for “How can the richness of expression be measured?”
• I think it is also important to effectively provide rich color information to visually impaired individuals.
• Q) I understand there are quite a few studies on providing image explanations for visually impaired individuals. How does this research differ from those? Did previous studies fail to provide accurate color information?

• Sangeun
• Page 3: What is the meaning of distorted images in your team?
• Clear
• Page 6: Max & Min value of model results
• Page 8: [Same opportunity]

• Jean
• In page 5, how did you classified the collected data?
• In page 6, how would you conclude about the performance?
• In page 8, the statistics seems too outdated (it’s from 2005). Were there any recent reports to compare with?

• Taehun
• When the model checks the image, does it also check the background for equality?
• If so, a preprocessing process would be needed to mask only the clothes and make them look the same.

• Dokyun
• In page 6, did you say that about 130 distorted images were restored? Then, how many images do you expect to need to properly learn the model?
• Second topic is interesting. Text maybe helpful for people who are color blindness, color weakness, or are visually impaired.