R-CNN/SPP-Net

R-CNN

• 2013
• region proposals with CNNs
• selective search
• extract region proposals(~2000개)
• Supervised pre-training : Caffe CNN library
• Domain-specific fine-tuning : VOC, N=20 / ILSVRC2013, N=200

SPP-net

• 2014
• spatial pyramid pooing
• convolutional features -> R-CNN : 2000번 / SPP-net : 1번 (R-CNN 보다 24~102배 더 빠름)
• image -> crop/wrap 을 안한다.
• image -> conv layers -> spatial pyramid pooing
• spatial pyramid pooing : 예) 1+4+16=21개
• SPP은 항상 고정된 사이즈가 나온다.
• Bag-of-words => spatial pyramid Matching(SPM) ==> SPP 로 확장(크기고정)

# %pip install -U selective-search

import selective_search

dir(selective_search)

['__builtins__',
 '__cached__',
 '__doc__',
 '__file__',
 '__loader__',
 '__name__',
 '__package__',
 '__path__',
 '__spec__',
 'box_filter',
 'core',
 'measure',
 'selective_search',
 'structure',
 'util']

from skimage.data import astronaut

data = astronaut()

import matplotlib.pyplot as plt

plt.figure(figsize=(10,10))
plt.imshow(data)

<matplotlib.image.AxesImage at 0x16c5defa0>

# %pip install -U mglearn

import mglearn

# selective_search 는 plot_agglomerative_algorithm 와 비슷
mglearn.plot_agglomerative.plot_agglomerative_algorithm()

selective_search.selective_search(data)

[(0, 0, 512, 512),
 (290, 367, 491, 512),
 (221, 180, 282, 249),
 (221, 217, 247, 246),
 (435, 122, 460, 274),
 (323, 206, 386, 264),
 (389, 19, 417, 85),
 (267, 27, 307, 68),
 (175, 194, 195, 243),
 (148, 425, 238, 462),
 (41, 75, 137, 218),
 (160, 427, 242, 512),
 (360, 237, 500, 458),
 (240, 308, 291, 512),
 (240, 308, 277, 512),
 (382, 0, 436, 186),
 (393, 0, 436, 186),
 (400, 0, 424, 80),
 (110, 241, 223, 417),
 (123, 260, 223, 417),
 (360, 276, 375, 348),
 (244, 308, 277, 512),
 (242, 180, 282, 249),
 (411, 29, 423, 80),
 (185, 144, 238, 182),
 (156, 100, 184, 148),
 (170, 427, 242, 512),
 (0, 0, 95, 512),
 (0, 0, 64, 160),
 (0, 118, 95, 512),
 (10, 0, 37, 129),
 (20, 20, 37, 129),
 (364, 237, 500, 458),
 (367, 237, 453, 360),
 (367, 278, 453, 360),
 (338, 417, 378, 456),
 (0, 118, 95, 419),
 (0, 118, 95, 370),
 (0, 118, 95, 275),
 (0, 147, 77, 275),
 (370, 275, 512, 484),
 (17, 0, 512, 512),
 (396, 0, 512, 285),
 (41, 75, 102, 210),
 (133, 348, 208, 424),
 (49, 420, 152, 509),
 (49, 420, 152, 499),
 (49, 445, 152, 499),
 (113, 126, 304, 319),
 (113, 132, 304, 319),
 (393, 237, 446, 292),
 (403, 237, 446, 292),
 (30, 0, 74, 86),
 (278, 342, 512, 512),
 (278, 352, 512, 512),
 (352, 352, 512, 512),
 (0, 226, 38, 370),
 (113, 132, 304, 277),
 (125, 132, 304, 277),
 (133, 164, 304, 277),
 (133, 215, 250, 273),
 (409, 0, 512, 285),
 (280, 342, 328, 388),
 (13, 118, 95, 237),
 (411, 117, 432, 181),
 (244, 322, 275, 512),
 (0, 300, 23, 411),
 (409, 19, 512, 285),
 (10, 147, 77, 275),
 (367, 290, 453, 360),
 (0, 478, 26, 512),
 (17, 0, 420, 512),
 (17, 199, 240, 512),
 (174, 278, 240, 441),
 (453, 52, 486, 103),
 (393, 0, 436, 185),
 (402, 0, 436, 185),
 (116, 126, 175, 167),
 (364, 357, 512, 512),
 (52, 0, 420, 512),
 (52, 0, 405, 269),
 (52, 0, 405, 251),
 (172, 77, 294, 191),
 (172, 77, 260, 177),
 (174, 390, 238, 441),
 (364, 344, 500, 458),
 (244, 334, 275, 512),
 (52, 0, 396, 251),
 (191, 0, 396, 251),
 (257, 0, 396, 251),
 (17, 199, 237, 512),
 (21, 329, 125, 424),
 (414, 19, 512, 285),
 (24, 431, 45, 482),
 (348, 43, 405, 243),
 (348, 100, 405, 243),
 (52, 0, 303, 169),
 (69, 420, 132, 465),
 (0, 226, 38, 334),
 (10, 147, 72, 257),
 (0, 147, 19, 237),
 (420, 19, 512, 285),
 (157, 43, 303, 158),
 (157, 43, 303, 130),
 (250, 43, 303, 130),
 (191, 0, 279, 31),
 (26, 329, 125, 424),
 (26, 329, 125, 409),
 (17, 205, 237, 512),
 (0, 226, 34, 329),
 (52, 0, 286, 169),
 (26, 452, 166, 512),
 (26, 452, 91, 512),
 (191, 248, 256, 264),
 (0, 226, 34, 323),
 (136, 226, 301, 319),
 (172, 362, 200, 390),
 (212, 218, 420, 512),
 (340, 218, 420, 331),
 (340, 240, 420, 331),
 (340, 254, 420, 331),
 (348, 104, 405, 243),
 (133, 153, 295, 269),
 (133, 153, 289, 269),
 (206, 217, 289, 269),
 (358, 104, 405, 243),
 (212, 234, 382, 512),
 (52, 0, 227, 169),
 (52, 0, 216, 169),
 (52, 0, 216, 155),
 (424, 19, 512, 285),
 (301, 178, 349, 238),
 (317, 183, 347, 238),
 (442, 19, 512, 282),
 (125, 132, 184, 227),
 (258, 217, 289, 268),
 (230, 0, 279, 22),
 (136, 230, 301, 319),
 (20, 223, 90, 306),
 (13, 119, 81, 237),
 (278, 354, 369, 512),
 (280, 342, 322, 388),
 (449, 358, 500, 458),
 (449, 358, 489, 405),
 (409, 180, 424, 222),
 (54, 0, 116, 30),
 (133, 153, 262, 260),
 (133, 153, 249, 255),
 (92, 141, 136, 169),
 (225, 234, 382, 512),
 (259, 217, 289, 268),
 (0, 0, 40, 160),
 (30, 177, 72, 257),
 (272, 349, 313, 512),
 (133, 168, 227, 255),
 (220, 88, 270, 158),
 (17, 205, 237, 481),
 (29, 205, 237, 481),
 (29, 205, 211, 469),
 (56, 205, 196, 344),
 (225, 328, 270, 512),
 (225, 336, 270, 512),
 (239, 336, 270, 512),
 (136, 230, 301, 302),
 (408, 421, 463, 484),
 (157, 51, 280, 115),
 (230, 234, 382, 372),
 (250, 234, 382, 372),
 (254, 234, 382, 372),
 (258, 234, 382, 372),
 (258, 237, 382, 372),
 (277, 237, 382, 372),
 (257, 0, 396, 231),
 (156, 153, 249, 246),
 (301, 178, 349, 220),
 (147, 188, 227, 255),
 (147, 188, 202, 255),
 (456, 19, 512, 272),
 (458, 19, 512, 126),
 (311, 0, 370, 33),
 (99, 389, 237, 481),
 (99, 389, 161, 481),
 (157, 55, 263, 115),
 (364, 357, 491, 488),
 (133, 168, 180, 203),
 (157, 66, 189, 115),
 (75, 459, 122, 497),
 (22, 11, 64, 121),
 (458, 19, 512, 57),
 (220, 88, 270, 150),
 (220, 101, 266, 150),
 (257, 154, 325, 231),
 (364, 357, 491, 466),
 (395, 357, 491, 466),
 (279, 237, 382, 372),
 (102, 483, 175, 512),
 (116, 139, 153, 167),
 (377, 104, 405, 243),
 (478, 454, 512, 512),
 (478, 466, 512, 512),
 (275, 388, 313, 512),
 (188, 55, 263, 86),
 (414, 114, 456, 269),
 (136, 230, 301, 300),
 (92, 141, 136, 164),
 (246, 404, 270, 512),
 (54, 0, 103, 26),
 (75, 0, 103, 26),
 (278, 387, 322, 512),
 (278, 387, 322, 503),
 (13, 119, 81, 235),
 (2, 239, 34, 323),
 (404, 407, 480, 488),
 (5, 239, 34, 301),
 (409, 407, 480, 486),
 (26, 329, 120, 409),
 (212, 468, 260, 512),
 (456, 96, 512, 272),
 (256, 338, 275, 512),
 (113, 180, 139, 244),
 (133, 348, 195, 424),
 (133, 348, 185, 424),
 (468, 37, 512, 126),
 (260, 0, 396, 211),
 (437, 27, 462, 205),
 (281, 182, 325, 231),
 (29, 268, 211, 469),
 (456, 96, 512, 265),
 (29, 268, 128, 375),
 (29, 268, 126, 375),
 (136, 230, 261, 300),
 (304, 354, 369, 406),
 (433, 407, 480, 486),
 (180, 191, 249, 246),
 (180, 196, 246, 246),
 (188, 61, 263, 86),
 (156, 153, 243, 219),
 (370, 275, 512, 482),
 (279, 387, 322, 488),
 (424, 421, 463, 484),
 (44, 268, 126, 375),
 (169, 254, 261, 300),
 (453, 52, 482, 96),
 (492, 473, 512, 512),
 (442, 246, 512, 282),
 (433, 407, 471, 446),
 (364, 344, 464, 385),
 (290, 387, 322, 479),
 (370, 290, 458, 357),
 (242, 186, 263, 249),
 (224, 502, 256, 512),
 (283, 499, 310, 512),
 (5, 499, 26, 512),
 (121, 491, 175, 512),
 (247, 490, 257, 510),
 (492, 488, 512, 512),
 (62, 484, 73, 509),
 (102, 483, 158, 512),
 (212, 481, 250, 512),
 (79, 480, 166, 512),
 (71, 480, 112, 509),
 (160, 478, 182, 499),
 (0, 478, 15, 512),
 (278, 474, 295, 503),
 (75, 474, 107, 497),
 (492, 473, 507, 504),
 (215, 471, 228, 502),
 (243, 468, 260, 501),
 (233, 467, 248, 494),
 (478, 466, 500, 512),
 (256, 466, 270, 512),
 (225, 466, 240, 496),
 (156, 465, 193, 512),
 (160, 463, 196, 512),
 (492, 462, 512, 508),
 (31, 460, 44, 489),
 (96, 459, 122, 481),
 (275, 458, 287, 512),
 (404, 456, 445, 488),
 (49, 456, 91, 494),
 (416, 455, 436, 473),
 (30, 455, 91, 512),
 (480, 454, 505, 502),
 (256, 452, 264, 481),
 (26, 452, 61, 495),
 (424, 451, 454, 484),
 (36, 451, 67, 478),
 (279, 447, 292, 488),
 (439, 445, 480, 486),
 (57, 445, 152, 499),
 (242, 444, 256, 475),
 (403, 442, 416, 480),
 (69, 442, 132, 465),
 (202, 438, 238, 476),
 (25, 437, 45, 482),
 (290, 432, 305, 479),
 (24, 431, 38, 475),
 (347, 429, 369, 459),
 (187, 429, 237, 444),
 (147, 428, 237, 468),
 (453, 427, 471, 446),
 (351, 427, 372, 462),
 (148, 426, 236, 462),
 (168, 425, 238, 444),
 (408, 424, 431, 463),
 (338, 424, 353, 456),
 (424, 421, 463, 464),
 (425, 421, 442, 433),
 (463, 420, 491, 466),
 (81, 420, 126, 451),
 (347, 417, 378, 456),
 (335, 415, 380, 471),
 (174, 415, 230, 441),
 (260, 413, 275, 512),
 (246, 411, 258, 444),
 (409, 409, 439, 437),
 (433, 407, 463, 436),
 (246, 404, 266, 466),
 (480, 403, 500, 458),
 (275, 401, 293, 449),
 (158, 400, 198, 422),
 (100, 400, 161, 481),
 (153, 397, 195, 415),
 (0, 396, 40, 512),
 (166, 394, 204, 424),
 (17, 394, 116, 456),
 (211, 390, 238, 428),
 (99, 389, 124, 420),
 (465, 388, 487, 420),
 (275, 388, 313, 473),
 (293, 387, 322, 442),
 (209, 387, 238, 424),
 (154, 387, 191, 411),
 (476, 383, 489, 405),
 (175, 382, 208, 419),
 (292, 381, 317, 410),
 (21, 381, 45, 419),
 (38, 380, 114, 424),
 (369, 379, 491, 512),
 (293, 375, 487, 512),
 (172, 374, 196, 390),
 (437, 373, 489, 448),
 (139, 373, 185, 398),
 (332, 369, 369, 385),
 (290, 367, 396, 512),
 (280, 367, 312, 423),
 (248, 366, 258, 418),
 (138, 366, 169, 386),
 (364, 365, 442, 394),
 (159, 364, 176, 378),
 (173, 362, 200, 388),
 (107, 362, 125, 393),
 (60, 362, 70, 388),
 (286, 361, 322, 388),
 (360, 359, 398, 373),
 (251, 359, 254, 401),
 (449, 358, 480, 398),
 (395, 357, 471, 432),
 (375, 357, 400, 369),
 (343, 354, 369, 379),
 (304, 354, 359, 406),
 (455, 353, 512, 484),
 (283, 353, 299, 374),
 (154, 353, 194, 367),
 (352, 352, 404, 367),
 (192, 352, 218, 417),
 (146, 352, 169, 366),
 (272, 349, 291, 453),
 (179, 349, 208, 399),
 (156, 349, 197, 362),
 (122, 349, 211, 469),
 (133, 348, 179, 424),
 (305, 346, 328, 376),
 (26, 346, 48, 384),
 (288, 345, 321, 367),
 (395, 344, 464, 385),
 (364, 344, 401, 354),
 (280, 342, 317, 368),
 (427, 339, 457, 361),
 (256, 338, 273, 428),
 (239, 336, 249, 472),
 (372, 335, 453, 360),
 (244, 334, 265, 383),
 (28, 333, 53, 374),
 (378, 331, 412, 341),
 (258, 331, 283, 359),
 (0, 331, 39, 419),
 (46, 329, 120, 409),
 (237, 328, 257, 444),
 (23, 328, 38, 370),
 (240, 325, 250, 350),
 (247, 322, 261, 350),
 (62, 313, 126, 375),
 (272, 312, 291, 344),
 (230, 312, 253, 331),
 (117, 312, 127, 360),
 (45, 312, 56, 337),
 (250, 308, 277, 342),
 (361, 303, 375, 348),
 (0, 301, 17, 343),
 (252, 300, 277, 312),
 (0, 299, 26, 338),
 (231, 297, 251, 318),
 (207, 297, 240, 319),
 (29, 295, 97, 373),
 (151, 294, 202, 341),
 (2, 294, 22, 323),
 (254, 293, 300, 339),
 (370, 292, 387, 336),
 (372, 290, 458, 357),
 (367, 290, 400, 345),
 (106, 289, 128, 349),
 (277, 283, 311, 349),
 (348, 282, 357, 321),
 (371, 281, 388, 291),
 (353, 279, 370, 331),
 (224, 279, 278, 302),
 (397, 278, 418, 294),
 (175, 278, 240, 427),
 (360, 276, 374, 333),
 (373, 275, 512, 482),
 (129, 275, 139, 343),
 (5, 275, 17, 301),
 (448, 274, 512, 286),
 (371, 272, 408, 291),
 (403, 271, 446, 292),
 (169, 269, 261, 300),
 (329, 268, 382, 362),
 (44, 268, 122, 363),
 (160, 266, 196, 305),
 (218, 264, 250, 273),
 (424, 261, 460, 285),
 (250, 260, 310, 312),
 (123, 260, 223, 407),
 (496, 259, 512, 272),
 (183, 259, 222, 273),
 (206, 258, 267, 269),
 (22, 258, 38, 334),
 (397, 257, 415, 270),
 (239, 256, 301, 288),
 (473, 255, 512, 274),
 (340, 254, 420, 290),
 (171, 254, 261, 290),
 (181, 252, 216, 268),
 (24, 252, 44, 313),
 (35, 251, 90, 306),
 (10, 251, 26, 329),
 (221, 250, 256, 264),
 (191, 248, 230, 263),
 (144, 248, 204, 284),
 (222, 247, 257, 260),
 (442, 246, 491, 282),
 (393, 243, 416, 260),
 (413, 241, 424, 269),
 (110, 241, 127, 266),
 (383, 240, 405, 264),
 (314, 240, 344, 294),
 (9, 239, 34, 277),
 (412, 237, 425, 276),
 (363, 237, 378, 248),
 (279, 237, 336, 372),
 (431, 234, 458, 268),
 (295, 234, 341, 283),
 (287, 234, 305, 280),
 (213, 233, 262, 260),
 (159, 233, 202, 255),
 (435, 232, 460, 274),
 (191, 231, 227, 252),
 (136, 230, 181, 270),
 (429, 227, 456, 236),
 (420, 227, 447, 273),
 (221, 227, 240, 246),
 (138, 226, 167, 249),
 (0, 226, 19, 302),
 (414, 224, 429, 269),
 (258, 223, 276, 258),
 (20, 223, 84, 278),
 (259, 221, 287, 268),
 (358, 220, 386, 238),
 (302, 219, 329, 251),
 (362, 218, 396, 272),
 (275, 217, 289, 248),
 (231, 217, 247, 235),
 (384, 215, 404, 243),
 (133, 215, 223, 271),
 (123, 214, 139, 244),
 (114, 214, 163, 284),
 (0, 214, 39, 268),
 (424, 209, 452, 224),
 (367, 208, 381, 218),
 (323, 206, 382, 264),
 (242, 206, 263, 249),
 (56, 205, 179, 344),
 (150, 204, 171, 236),
 (317, 201, 347, 238),
 (287, 199, 325, 231),
 (87, 199, 122, 239),
 (215, 198, 246, 235),
 (175, 198, 187, 230),
 (180, 196, 229, 246),
 (183, 196, 195, 227),
 (175, 194, 190, 243),
 (412, 191, 424, 222),
 (267, 191, 295, 242),
 (229, 191, 249, 217),
 (125, 190, 143, 232),
 (147, 188, 177, 236),
 (419, 187, 451, 224),
 (243, 186, 262, 241),
 (71, 185, 79, 215),
 (11, 184, 22, 238),
 (318, 183, 336, 214),
 (281, 182, 317, 211),
 (151, 182, 180, 198),
 (409, 180, 420, 208),
 (301, 180, 327, 220),
 (254, 180, 282, 234),
 (113, 180, 128, 221),
 (37, 180, 72, 257),
 (186, 179, 243, 217),
 (323, 178, 349, 210),
 (419, 177, 453, 223),
 (137, 177, 174, 244),
 (30, 177, 49, 202),
 (163, 175, 182, 192),
 (18, 174, 77, 275),
 (154, 171, 164, 190),
 (138, 168, 162, 195),
 (133, 168, 158, 203),
 (226, 164, 304, 277),
 (200, 164, 238, 180),
 (455, 163, 467, 272),
 (87, 158, 137, 218),
 (348, 155, 381, 225),
 (257, 154, 314, 196),
 (140, 154, 180, 177),
 (64, 154, 80, 211),
 (156, 153, 188, 219),
 (97, 153, 114, 169),
 (0, 153, 19, 237),
 (116, 152, 136, 167),
 (125, 149, 184, 227),
 (63, 147, 102, 210),
 (10, 147, 37, 185),
 (0, 147, 14, 172),
 (275, 144, 295, 164),
 (185, 144, 217, 182),
 (92, 144, 126, 164),
 (243, 142, 254, 158),
 (113, 141, 136, 156),
 (260, 140, 290, 162),
 (413, 139, 432, 181),
 (202, 139, 234, 155),
 (132, 139, 153, 157),
 (132, 132, 182, 167),
 (15, 132, 31, 166),
 (27, 130, 78, 237),
 (152, 126, 175, 142),
 (170, 124, 184, 148),
 (0, 124, 13, 158),
 (399, 123, 408, 183),
 (492, 122, 512, 162),
 (446, 122, 457, 235),
 (185, 121, 251, 174),
 (358, 119, 398, 224),
 (40, 118, 95, 230),
 (411, 117, 418, 156),
 (122, 116, 150, 139),
 (40, 116, 74, 158),
 (427, 114, 456, 228),
 (457, 110, 512, 265),
 (487, 109, 508, 137),
 (187, 108, 294, 191),
 (377, 104, 405, 222),
 (273, 104, 300, 138),
 (129, 104, 167, 144),
 (235, 102, 266, 150),
 (220, 101, 251, 144),
 (380, 100, 401, 113),
 (156, 100, 181, 130),
 (0, 100, 9, 124),
 (243, 99, 259, 108),
 (456, 96, 484, 125),
 (194, 96, 216, 106),
 (410, 91, 435, 186),
 (496, 88, 512, 123),
 (466, 88, 486, 103),
 (229, 88, 270, 113),
 (396, 85, 429, 252),
 (134, 80, 157, 109),
 (437, 78, 454, 205),
 (197, 78, 252, 88),
 (157, 78, 183, 115),
 (393, 76, 419, 135),
 (453, 75, 474, 122),
 (41, 75, 85, 174),
 (236, 68, 263, 85),
 (177, 66, 189, 85),
 (454, 63, 482, 96),
 (188, 61, 262, 86),
 (427, 59, 433, 89),
 (399, 55, 409, 82),
 (197, 55, 221, 69),
 (27, 54, 48, 78),
 (453, 52, 475, 75),
 (262, 51, 280, 68),
 (28, 50, 64, 121),
 (250, 49, 303, 130),
 (0, 48, 40, 160),
 (296, 45, 307, 68),
 (379, 43, 402, 104),
 (280, 43, 297, 73),
 (23, 43, 37, 129),
 (436, 40, 449, 137),
 (475, 39, 494, 57),
 (189, 33, 200, 52),
 (411, 31, 417, 57),
 (402, 31, 422, 83),
 (425, 28, 437, 188),
 (438, 27, 462, 91),
 (267, 27, 301, 54),
 (153, 24, 196, 64),
 (0, 21, 5, 51),
 (190, 20, 221, 47),
 (22, 20, 31, 55),
 (20, 20, 25, 73),
 (458, 19, 512, 56),
 (153, 15, 286, 104),
 (269, 12, 301, 32),
 (22, 11, 54, 72),
 (262, 10, 279, 29),
 (44, 9, 74, 86),
 (75, 3, 96, 26),
 (26, 2, 49, 29),
 (423, 0, 512, 55),
 (411, 0, 436, 185),
 (400, 0, 424, 58),
 (382, 0, 406, 74),
 (344, 0, 370, 14),
 (311, 0, 369, 33),
 (266, 0, 396, 211),
 (267, 0, 311, 21),
 (234, 0, 279, 22),
 (230, 0, 263, 16),
 (191, 0, 235, 31),
 (206, 0, 227, 19),
 (158, 0, 206, 19),
 (135, 0, 167, 16),
 (89, 0, 116, 30),
 (81, 0, 103, 17),
 (54, 0, 82, 23),
 (30, 0, 66, 24),
 (22, 0, 31, 14),
 (10, 0, 23, 67),
 (0, 0, 16, 69),
 (0, 0, 13, 17)]

ss = selective_search.selective_search(data, mode='fast')

len(ss)

3056

box = selective_search.box_filter(ss, topN=50)

import matplotlib.patches as pt

fig, ax = plt.subplots(figsize=(6,6))
ax.imshow(data)
for x1, y1, x2, y2 in box:
    ax.add_patch(pt.Rectangle((x1,y1), (x2-x1), (y2-y1), fill=None, edgecolor='red', linewidth=2))

import selectivesearch

t = selectivesearch.selective_search(data)

len(t)

2

#[i['rect'] for i in t[1]]

# %pip install -U seaborn

import seaborn as sns

tips = sns.load_dataset('tips')

tips # 머신러닝 데이터는 모든 샘플의 크기(컬럼)가 일정 해야 함
     # R-CNN 은 SVM 에 넣기 위해 크기 맞추기 위해 warped region을 하고 CNN features 를 함

	total_bill	tip	sex	smoker	day	time	size
0	16.99	1.01	Female	No	Sun	Dinner	2
1	10.34	1.66	Male	No	Sun	Dinner	3
2	21.01	3.50	Male	No	Sun	Dinner	3
3	23.68	3.31	Male	No	Sun	Dinner	2
4	24.59	3.61	Female	No	Sun	Dinner	4
...	...	...	...	...	...	...	...
239	29.03	5.92	Male	No	Sat	Dinner	3
240	27.18	2.00	Female	Yes	Sat	Dinner	2
241	22.67	2.00	Male	Yes	Sat	Dinner	2
242	17.82	1.75	Male	No	Sat	Dinner	2
243	18.78	3.00	Female	No	Thur	Dinner	2

244 rows × 7 columns

R-CNN

• 2013
• region proposals with CNNs
• selective search
• extract region proposals(~2000개)
• Supervised pre-training : Caffe CNN library
• Domain-specific fine-tuning : VOC, N=20 / ILSVRC2013, N=200

SPP-net

• 2014
• spatial pyramid pooing
• convolutional features -> R-CNN : 2000번 / SPP-net : 1번 (R-CNN 보다 24~102배 더 빠름)
• image -> crop/wrap 을 안한다.
• image -> conv layers -> spatial pyramid pooing
• spatial pyramid pooing : 예) 1+4+16=21개
• SPP은 항상 고정된 사이즈가 나온다.
• Bag-of-words => spatial pyramid Matching(SPM) ==> SPP 로 확장(크기고정)