๋ฐ˜์‘ํ˜•

๐ŸŒŒ Deep Learning 50

Nested cross validation

ํ”ํžˆ Cross validation์€ Train data์™€ Validation data๋ฅผ, ํ˜น์€ Train data์™€ Test data๋ฅผ kํšŒ ๊ฐ๊ฐ ๋‹ค๋ฅธ ๋ฐฉ๋ฒ•์œผ๋กœ ๋ถ„ํ• ํ•˜์—ฌ ๋ชจ๋ธ ์„ฑ๋Šฅ์„ ๊ฒ€์ฆํ•˜๋Š” ๋ฐฉ๋ฒ•์ด๋‹ค. ์•„๋ž˜ ๊ทธ๋ฆผ์€ 5-fold Train-Test cross validation์„ ์„ค๋ช…ํ•˜๊ณ  ์žˆ๋‹ค. Train data์™€ Validation data์˜ ๋ถ„ํ• ์— cross validation์„ ์‚ฌ์šฉํ•œ๋‹ค๋ฉด, ๋‹ค์–‘ํ•œ validation data๋ฅผ ์ด์šฉํ•ด ์ตœ์ ์˜ model parameter์„ ํƒ์ƒ‰ํ•˜๊ธฐ ์œ„ํ•จ์ด๊ณ , Train data์™€ Test data์˜ ๋ถ„ํ• ์— cross validation์„ ์‚ฌ์šฉํ•œ๋‹ค๋ฉด, ๋‹ค์–‘ํ•œ test data์— ๋Œ€ํ•ด model์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•จ์ด๋‹ค. Nested Cross validation์€ ..

[๋”ฅ๋Ÿฌ๋‹ ๋…ผ๋ฌธ๋ฆฌ๋ทฐ] Rethinking the Truly Unsupervised Image-to-Image Translation (TUNIT) (ICCV 2021)

2021 ICCV์— Accept๋œ ๋…ผ๋ฌธ์ธ "Rethinking the Truly Unsupervised Image-to-Image Translation"์„ ์ •๋ฆฌํ•œ ๊ธ€์ž…๋‹ˆ๋‹ค. Naver CLOVA AI์—์„œ ์ž‘์„ฑ๋œ ๋…ผ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด์ „๊นŒ์ง€์˜ Unsupervised model (cycleGAN ๋“ฑ)์€ ์‚ฌ์‹ค Semi-supervised ๋ชจ๋ธ์ด๋ผ๊ณ  ํ•ด์•ผ ํ•œ๋‹ค๊ณ  ์–˜๊ธฐํ•˜๋ฉฐ, Data collection(labeling)์ด ํ•„์š”ํ•˜์ง€ ์•Š์€ Truly unsupervised model์ธ TUNIT์„ ์ œ์•ˆํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋…ผ๋ฌธ ๋งํฌ: https://arxiv.org/pdf/2006.06500 Official code: https://github.com/clovaai/tunit 1. Levels of Supervision in G..

[๋”ฅ๋Ÿฌ๋‹ ๋…ผ๋ฌธ๋ฆฌ๋ทฐ] Loss Functions for Image Restoration with Neural Networks (IEEE TMI 2016)

2016๋…„ IEEE TMI์— ๊ฐœ์ œ๋œ ๋…ผ๋ฌธ์ธ "Loss Functions for Image Restoration with Neural Networks"๋ฅผ ์ •๋ฆฌํ•œ ๊ธ€์ด๋‹ค. Super-resolution, artifact removal, denoising ๋“ฑ Image restoration task์—์„œ ์“ฐ์ด๋Š” Loss function์— ๋Œ€ํ•ด ๋ถ„์„ํ–ˆ๊ณ , Image restoration task์—์„œ ์ข‹์€ ์„ฑ๋Šฅ์„ ๋‚ผ ์ˆ˜ ์žˆ๋Š” ์ƒˆ๋กœ์šด, ๋ฏธ๋ถ„๊ฐ€๋Šฅํ•œ loss function์„ ์ œ์•ˆํ–ˆ๋‹ค. ๊ฝค ์˜›๋‚  ๋…ผ๋ฌธ์ด์ง€๋งŒ, ์ฝ๊ธฐ ์‰ฝ๊ณ  ์œ ๋ช…ํ•œ ๋…ผ๋ฌธ์ด๋ผ ์ •๋ฆฌํ•ด ๋ณด๋ ค๊ณ  ํ•œ๋‹ค. 1. Background Image restoration Image restoration์ด๋ž€, denoising, deblurring, demosaicking, sup..

[PyTorch Implementation] PyTorch๋กœ ๊ตฌํ˜„ํ•œ cycleGAN์˜ loss ๋ถ€๋ถ„ ์„ค๋ช…

cycleGAN ๋…ผ๋ฌธ์—์„œ๋Š” cycleGAN์˜ Loss function์„ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ •์˜ํ•˜๊ณ  ์žˆ๋‹ค. (identity loss๋Š” optional) ์ด ์ˆ˜์‹์€ ์ฝ”๋“œ๋กœ ๋ณผ ๋•Œ ํ›จ์”ฌ ๊ฐ„๋‹จํ•ด์ง„๋‹ค. ๋‚˜๋Š” PyTorch๋ฅผ ์ด์šฉํ•ด cycleGAN์˜ loss ๋ถ€๋ถ„์„ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ž‘์„ฑํ–ˆ๋‹ค. ์ฝ”๋“œ๋Š” ์ €์ž๊ฐ€ ๊ณต๊ฐœํ•œ official code์™€ simplified version(unofficial)์„ ์ฐธ๊ณ ํ–ˆ๋‹ค. ์ฒ˜์Œ ๊ณต๋ถ€ํ•  ๋•Œ๋Š” simplified version์„ ํ†ตํ•ด ์ „์ฒด ๊ตฌ์กฐ๋ฅผ ์ดํ•ดํ•œ ํ›„, official code๋ฅผ ์ฐธ๊ณ ํ•˜๋ฉด ์ดํ•ดํ•˜๊ธฐ ํŽธํ•˜๋‹ค. (*cycleGAN์˜ detail์ด๋‚˜, ์ด๋ก ์ ์ธ ๋ถ€๋ถ„์€ ์ด ๊ธ€์—์„œ๋Š” ์„ค๋ช…ํ•˜์ง€ ์•Š๊ธฐ๋กœ ํ•œ๋‹ค.) ์šฐ์„ , ๊ฐ stage์˜ loss๋ฅผ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์‹์œผ๋กœ ์ •๋ฆฌํ•  ์ˆ˜ ์žˆ๋‹ค. ์ฐธ๊ณ ๋กœ, realA..

EdgesCats

pix2pix๋ฅผ ์ด์šฉํ•ด ๋งŒ๋“  ๊ทธ๋ฆผ->์‚ฌ์ง„ ๋ณ€ํ™˜ ํˆด์˜ ์ผ์ข…์ด๋‹ค. ๋งํฌ : https://affinelayer.com/pixsrv/ Image-to-Image Demo - Affine Layer Interactive Image Translation with pix2pix-tensorflow The pix2pix model works by training on pairs of images such as building facade labels to building facades, and then attempts to generate the corresponding output image from any input image you give it. affinelayer.com ๋งˆ์šฐ์Šค๋กœ ๊ณ ์–‘์ด ๊ทธ๋ฆผ์„ ๊ทธ๋ฆฌ๋ฉด ๊ณ ์–‘..

[Overview] Attention ์ •๋ฆฌ - (2) seq2seq, +attention

์ˆœ์„œ: (1) LSTM (2) seq2seq, +attention (3) Show, Attend and Tell Reference: Visualization of seq2seqmodel Visualizing A Neural Machine Translation Model (Mechanics of Seq2seq Models With Attention) Translations: Chinese (Simplified), Japanese, Korean, Russian, Turkish Watch: MIT’s Deep Learning State of the Art lecture referencing this post May 25th update: New graphics (RNN animation, word embedd..

[Overview] Attention ์ •๋ฆฌ - (1) LSTM

์ˆœ์„œ: (1) LSTM (2) seq2seq, +attention (3) Show, Attend and Tell reference: colah.github.io/posts/2015-08-Understanding-LSTMs/ Recurrent Neural Network ๊ธฐ๋ณธ์ ์ธ RNN์˜ ๊ตฌ์กฐ๋Š” ์œ„์™€ ๊ฐ™๋‹ค. ์ด์ „์˜ state๋ฅผ ํ•จ๊ป˜ input์œผ๋กœ ์ฃผ์–ด ์ด์ „ input๊ณผ์˜ ์—ฐ๊ด€์„ฑ์„ ํ•จ๊ป˜ ํ•™์Šตํ•ด ๋‚˜๊ฐ„๋‹ค. ํ•˜์ง€๋งŒ ์ด๋Ÿฌํ•œ ๊ตฌ์กฐ๋Š”, input์˜ ๊ธธ์ด๊ฐ€ ๊ธธ์–ด์งˆ์ˆ˜๋ก, ๋„คํŠธ์›Œํฌ์˜ ๋’ท๋ถ€๋ถ„์œผ๋กœ ๊ฐˆ ์ˆ˜๋ก ์•ž ๋ถ€๋ถ„์˜ ์ •๋ณด๋ฅผ ์žŠ์–ด๋ฒ„๋ฆฌ๋Š” ๋ฌธ์ œ์ ์ด ์žˆ๋‹ค. LSTM์€ ์ด๋Ÿฌํ•œ ๋ฌธ์ œ์ ์„ ๊ทน๋ณตํ•˜๊ธฐ ์œ„ํ•ด ์ œ์‹œ๋˜์—ˆ๋‹ค. Long Short Term Memory ์œ„๋Š” ๊ฐ„๋‹จํ•œ RNN์˜ ๊ตฌ์กฐ, ์•„๋ž˜๋Š” LSTM์˜ ๊ตฌ์กฐ์ด๋‹ค. ๊ฐ ๊ธฐํ˜ธ์˜ ์˜๋ฏธ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค..

MSE Loss (L2 Loss) vs. MAE Loss (L1 Loss)

heartbeat.fritz.ai/5-regression-loss-functions-all-machine-learners-should-know-4fb140e9d4b0 ์˜ ์ผ๋ถ€๋ฅผ ๋ฒˆ์—ญ & ์š”์•ฝํ•œ ๊ธ€ MSE Loss and MAE Loss MSE(Mean Squared Loss, L2 Loss)๋Š” Error์˜ ์ œ๊ณฑ์˜ ํ‰๊ท ์„ ๋‚ธ ๊ฐ’์ž…๋‹ˆ๋‹ค. ์ˆ˜์‹์œผ๋กœ ๋‚˜ํƒ€๋‚ด๋ฉด ์•„๋ž˜์™€ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์€ True value๋ฅผ 100์ด๋ผ๊ณ  ํ–ˆ์„ ๋•Œ Predicted value๋ฅผ -10,000์—์„œ 10,000๊นŒ์ง€ ๋ณ€ํ™”์‹œ์ผœ ๊ฐ€๋ฉฐ ๊ทธ๋ฆฐ ๊ทธ๋ž˜ํ”„์ž…๋‹ˆ๋‹ค. x์ถ•์€ Predicted value, y์ถ•์€ MSE Loss์˜ ๊ฐ’์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. MAE(Mean Absolute Loss, L1 Loss)๋Š” Error์˜ ์ ˆ๋Œ€๊ฐ’์˜ ํ‰๊ท ์„ ๋‚ธ ๊ฐ’์ž…๋‹ˆ๋‹ค. ์ˆ˜์‹์œผ๋กœ..

[Overview] YOLO ๊ณ„์—ด Object Detection ์ •๋ฆฌ - (1) YOLO

์ˆœ์„œ: (1) YOLO (2016) (2) YOLOv2 (3) YOLOv3 (4) YOLOv4 YOLO (2016) Redmon, Joseph, et al. "You only look once: Unified, real-time object detection." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016. Paper: www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Redmon_You_Only_Look_CVPR_2016_paper.pdf Official code: pjreddie.com/darknet/yolo/ ๋…ผ๋ฌธ์—์„œ ์ œ์‹œํ•œ ๋ชจ๋ธ ๊ตฌ์กฐ๋Š” ์œ„ ..

[Dataset] People-Art ๋ฐ์ดํ„ฐ์…‹

People-Art ๋ฐ์ดํ„ฐ์…‹์€ ์„œ๋กœ ๋‹ค๋ฅธ style์˜ ์ด๋ฏธ์ง€์—์„œ ์‚ฌ๋žŒ์„ ์ฐพ๋Š” object detection ์šฉ ๋ฐ์ดํ„ฐ์…‹์ž…๋‹ˆ๋‹ค. Dataset download: github.com/BathVisArtData/PeopleArt Paper: arxiv.org/pdf/1610.08871v1.pdf Dataset Description ์„œ๋กœ ๋‹ค๋ฅธ style์˜ ์ด๋ฏธ์ง€์— ๋Œ€ํ•ด ์ž˜ ์ธ์‹ํ•˜๋„๋ก ํ•˜๋Š” ๋ฌธ์ œ๋ฅผ Cross-depiction problem์ด๋ผ๊ณ  ์ •์˜ํ•ฉ๋‹ˆ๋‹ค. People-Art ๋ฐ์ดํ„ฐ์…‹์€ ์‚ฌ์ง„, ๋งŒํ™” ๊ทธ๋ฆฌ๊ณ  41๊ฐœ์˜ ํ™”ํ’์œผ๋กœ ๊ตฌ์„ฑ๋œ ์ด 43๊ฐœ์˜ style๋ฅผ ๊ฐ€์ง‘๋‹ˆ๋‹ค. 41๊ฐœ์˜ ํ™”ํ’ ์ด๋ฏธ์ง€๋Š” WikiArt.org์—์„œ, ์‚ฌ์ง„์€ PASCAL VOC 2012 ๋ฐ์ดํ„ฐ์…‹์—์„œ, ๋งŒํ™” ์ด๋ฏธ์ง€๋Š” Google์—์„œ ๊ฐ€์ ธ์™”๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค...

๋ฐ˜์‘ํ˜•