pal4vst_food

AI 생성 μŒμ‹ μ΄λ―Έμ§€μ—μ„œ μ–΄μƒ‰ν•œ μ˜μ—­(μ•„ν‹°νŒ©νŠΈ) 후보λ₯Ό ν”½μ…€ λ‹¨μœ„λ‘œ μΆ”μ •ν•˜λŠ” λΆ„ν•  λͺ¨λΈμž…λ‹ˆλ‹€. Lingzhi/PAL4VST(MIT, ICCV 2023)λ₯Ό 직접 λΌλ²¨λ§ν•œ AI 생성 μŒμ‹ μ΄λ―Έμ§€λ‘œ λ―Έμ„Έμ‘°μ •ν–ˆμŠ΅λ‹ˆλ‹€.

  • ꡬ쑰: Swin-B + UperNet (transformers.UperNetForSemanticSegmentation), 2클래슀 (0 background, 1 bad_region)
  • ν•™μŠ΅: AI 생성 μŒμ‹ 이미지 12μž₯의 μ–΄μƒ‰ν•œ μ˜μ—­ 마슀크(직접 라벨링), 1250 iter
  • μ „μ²˜λ¦¬: λΉ„μœ¨ μœ μ§€λ‘œ κΈ΄ λ³€ 512, RGB 평균 (123.675, 116.28, 103.53) / ν‘œμ€€νŽΈμ°¨ (58.395, 57.12, 57.375)

버전

컀밋 λ‚΄μš©
da449b1 v1
deed190 v2: 떑볢이 이미지 라벨 μˆ˜μ • ν›„ μž¬ν•™μŠ΅ (ν•™μŠ΅ μ„€μ • 동일)

μ‚¬μš©

from transformers import UperNetForSemanticSegmentation
model = UperNetForSemanticSegmentation.from_pretrained("hs4167/pal4vst_food", revision="deed190ca7f734adaa275624744143c005f2e3e8")

ν•œκ³„

  • ν•™μŠ΅ 데이터가 12μž₯뿐이고 μ‹€μ œ 사진(였λ₯˜ μ—†μŒ)은 ν•™μŠ΅μ— μ“°μ§€ μ•Šμ•„, μ‹€μ œ μŒμ‹ μ‚¬μ§„μ—μ„œλ„ 넓은 μ˜μ—­μ„ ν›„λ³΄λ‘œ μž‘λŠ” μ˜€νƒμ΄ μžˆμŠ΅λ‹ˆλ‹€.
  • κ²°κ³ΌλŠ” 였λ₯˜ ν›„λ³΄μ˜ 좔정이며, 이미지가 AI둜 μƒμ„±λ˜μ—ˆλŠ”μ§€ νŒλ³„ν•˜κ±°λ‚˜ ν’ˆμ§ˆμ„ λ³΄μ¦ν•˜μ§€ μ•ŠμŠ΅λ‹ˆλ‹€.

λΌμ΄μ„ μŠ€

Apache-2.0 (LICENSE). 원본 PAL4VST κ°€μ€‘μΉ˜μ˜ MIT κ³ μ§€λŠ” NOTICE에 μžˆμŠ΅λ‹ˆλ‹€.

Downloads last month
47
Safetensors
Model size
0.1B params
Tensor type
F32
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for hs4167/pal4vst_food

Base model

Lingzhi/PAL4VST
Finetuned
(1)
this model