๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ

์ „์ฒด ๊ธ€72

Multi-GPU Model์—์„œ h5(hdf5)๋ฅผ ๋กœ๋“œํ•˜์ง€ ๋ชปํ•˜๋Š” ๋ฌธ์ œ ๋ฌธ์ œ ๊ฐœ์š” Multi-GPU Model๋กœ ํ•™์Šต์‹œํ‚ค๊ณ  h5(hdf5)๋กœ ์ €์žฅํ•ด๋‘” ๋‹ค์Œ, ๋‚˜์ค‘์— ๋ถˆ๋Ÿฌ์˜ฌ ๋•Œ layers ์ˆ˜๊ฐ€ ๋งž์ง€ ์•Š๋‹ค๊ณ  ์—๋Ÿฌ๊ฐ€ ๋‚จ ์ด๋ฏธ ๋งŽ์ด ์•Œ๋ ค์ง„ ๋ฒ„๊ทธ ํ•ด๊ฒฐ์ฑ… ๊ฐœ์š” Multi-GPU Model๋กœ ํ•™์Šตํ•œ weight๋ฅผ (Multi-GPU๋กœ compileํ•˜์ง€ ์•Š์€) Base Model์— setํ•œ ๋‹ค์Œ h5(hdf5)๋กœ ์ €์žฅ Epoch๊ฐ€ ๋ชจ๋‘ ๋๋‚œ ๋‹ค์Œ model.save()๋Š” ๋Œ€๋žต ์•„๋ž˜์™€ ๊ฐ™์ด ํ•˜๋ฉด ๋˜์ง€๋งŒ... ... parallel_model, base_model = get_compiled_model(...) parallel_model.fit(...) base_model.set_weights(parallel_model.get_weights()) base_model.save(filepath).. 2019. 8. 13.
Variable Scope ๊ณต๋ถ€ ์ฝ์€ ์ž๋ฃŒ: https://tensorflowkorea.gitbooks.io/tensorflow-kr/content/g3doc/how_tos/variable_scope/ ์ฃผ์š” ํ•จ์ˆ˜ - tf.get_variable(, , ): ์ž…๋ ฅ๋œ ์ด๋ฆ„์˜ ๋ณ€์ˆ˜๋ฅผ ์ƒ์„ฑํ•˜๊ฑฐ๋‚˜ ๋ฐ˜ํ™˜ - tf.variable_scope(): tf.get_variable()์— ์ „๋‹ฌ๋œ ์ด๋ฆ„์˜ ๋„ค์ž„์ŠคํŽ˜์ด์Šค๋ฅผ ๊ด€๋ฆฌ (with์™€ ํ•จ๊ป˜ ์“ฐ์ž„) - tf.get_variable_scope(): ํ˜„์žฌ variable scope๋ฅผ ๋ฐ˜ํ™˜ tf.get_variable() - tf.Variable()์ฒ˜๋Ÿผ ์ง์ ‘ ๊ฐ’์„ ์ „๋‹ฌํ•˜์ง€ ์•Š๊ณ  Initializer๋ฅผ ์‚ฌ์šฉ tf.constant_initializer(value): ์ œ๊ณต๋œ ๊ฐ’์œผ๋กœ ๋ชจ๋“  ๊ฒƒ์„ ์ดˆ๊ธฐํ™” tf.random_u.. 2019. 8. 13.
Attention ๊ณต๋ถ€ ์ฝ์€ ์ž๋ฃŒ: https://youtu.be/iDulhoQ2pro https://hulk89.github.io/neural%20machine%20translation/2017/04/04/attention-mechanism/ Image Captioning https://github.com/zzsza/Deep_Learning_starting_with_the_latest_papers/blob/master/Lecture_Note/03.%20CNN%20Application/12.Image-Captioning.md https://greeksharifa.github.io/computer%20vision/2019/04/17/Visual-Question-Answering/ - Image captioning: ์ด๋ฏธ์ง€๋ฅผ ์ฃผ๋ฉด.. 2019. 8. 12.
Attention Is All You Need ๊ณต๋ถ€ BLUE Score https://donghwa-kim.github.io/BLEU.html ์š”์•ฝ: X label๊ณผ Y label์ด ๋ชจ๋‘ ๋‹จ์–ด๋“ค์˜ ์‹œํ€€์Šค (์ฆ‰ ๋ฌธ์žฅ) ์ผ ๋•Œ ์‚ฌ์šฉํ•˜๋Š” ์„ฑ๊ณผ ์ง€ํ‘œ. Precision, Clipping, Brevity penalty ๊ฐœ๋…์„ ์‚ฌ์šฉํ•œ๋‹ค. ์œ„ ๊ธ€์— ์ดํ•ดํ•˜๊ธฐ ์ข‹๊ฒŒ ์ž˜ ์ •๋ฆฌ๊ฐ€ ๋˜์–ด์žˆ๋‹ค. n-gram์€ ์—ฐ์†๋œ n๊ฐœ์˜ ํ† ํฐ(๋‹จ์–ด)์ด๋‹ค. Attention ๊ด€๋ จ ์ž๋ฃŒ http://www.modulabs.co.kr/?module=file&act=procFileDownload&file_srl=20068&sid=c11da8e0ec9d00ea4fdc828fa2bff90a&module_srl=18361 ๋ชจ๋‘์˜ ์—ฐ๊ตฌ์†Œ - ๊ฐ„๋‹จํ•œ ๋ฐœํ‘œ ์ž๋ฃŒ https://jalammar.github.io/.. 2019. 8. 9.
ResNet ๊ณต๋ถ€ ์ฝ์€ ์ž๋ฃŒ: https://dnddnjs.github.io/cifar10/2018/10/09/resnet/ Fine tuning https://eehoeskrap.tistory.com/186 Parameter & Hyperparameter https://datascience.stackexchange.com/questions/17635/model-parameters-hyper-parameters-of-neural-network-their-tuning-in-training Neural network์˜ Parameter๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ Connection์˜ Weights๋ฅผ ๋งํ•œ๋‹ค. ์ด Parameters๋Š” Training stage์—์„œ ํ•™์Šต๋œ๋‹ค. ๊ทธ๋ž˜์„œ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ž์ฒด (๊ทธ๋ฆฌ๊ณ  ์ธํ’‹ ๋ฐ์ดํ„ฐ๋Š”) ์ด Parameter.. 2019. 7. 31.
Simultaneous Recognition of Horizontal and Vertical Text in Natural Images (2018.12) DEM (Directional Encoding Mask) - Text์˜ Directional information์ด ๋‹ด๊ฒจ์žˆ๋Š”, ์ธํ’‹ ์ด๋ฏธ์ง€์™€ ์‚ฌ์ด์ฆˆ๊ฐ€ ๊ฐ™์€ 2D ์ด๋ฏธ์ง€ - ๊ฐ€๋กœ ํ…์ŠคํŠธ์˜ ๊ฒฝ์šฐ: cos(0.5 * normalized_width * pi) - ์„ธ๋กœ ํ…์ŠคํŠธ์˜ ๊ฒฝ์šฐ: sin(0.5 * normalized_width * pi) - ์ธํ’‹ ์ด๋ฏธ์ง€์™€ Concatenate ๋˜์–ด CNN์˜ ์ธํ’‹์œผ๋กœ ๋“ค์–ด๊ฐ€๋Š” ํ˜•ํƒœ SAN (Selective Attention Network) - Text direction์— ๋”ฐ๋ผ ๋‹ค๋ฅด๊ฒŒ ํ–‰๋™ํ•˜๋Š” ๋‘ ๊ฐ€์ง€ Attention mask๋ฅผ ๊ฐ€์ง 2019. 7. 29.