search

OSVAI / GridConv

The official project website of "3D Human Pose Lifting with Grid Convolution" (GridConv for short, oral in AAAI 2023)
Apache License 2.0
34 stars 1 forks source link

Maybe some slight errors in code #4

Closed cocoshe closed 10 months ago

cocoshe commented 11 months ago

Thanks for your great work! I find some error here when I run the code, and here are the errors and solutions (maybe only for me?)

modify https://github.com/OSVAI/GridConv/blob/2b74889d379e2c94160a36eb28398a6368afbcc6/src/dataset/human36m.py#L33-L34

to

cur_data_2d = torch.load(os.path.join(data_path, data_2d_file), encoding='Latin1')
cur_data_3d = torch.load(os.path.join(data_path, data_3d_file), encoding='Latin1')

and https://github.com/OSVAI/GridConv/blob/2b74889d379e2c94160a36eb28398a6368afbcc6/src/network/gridconv.py#L89-L93

to

self.donut_conv = nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size,
                          padding=[kernel_size // 2, kernel_size // 2], padding_mode='circular', bias=bias)
self.tablet_conv = nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size,
                           padding=[kernel_size // 2, kernel_size // 2], padding_mode='zeros', bias=bias)
kyang-06 commented 10 months ago

Yes, it depends on python and pytorch version. The released code is compatible for py27 & pytorch1.4, while your revision works for pytorch > 1.8 (if I remember correctly). Thanks for the complementation, maybe useful for others :)

cocoshe commented 10 months ago

Yes, it depends on python and pytorch version. The released code is compatible for py27 & pytorch1.4, while your revision works for pytorch > 1.8 (if I remember correctly). Thanks for the complementation, maybe useful for others :)

Tks for your reply~, but would the version problem or the encoding='latin1' influence the result? I try to reproduce the result with py3 and some latest pytorch but somehow seems failed (more detail in my another issue here). Also I tried the environment.yml to build the whole conda env with py2, but the conda can't index some of the dependencies in yaml(pytorch, torchvision, torchaudio, for example, maybe they are out of date)

So is there any training log left? Or any suggestions on producing the result? Really appreciate~

kyang-06 commented 10 months ago

encoding has no influence on the performance. You can find my suggestion in your referred issue. Here is a training log for hrnet_dgridconv_AutoGrids:

epoch lr loss_train loss_test err_test err_x err_y err_z
1 0.0010 0.0385 0.0488 296.1125 75.8109 199.3963 143.6313
2 0.0010 0.0439 0.0455 265.8105 72.4859 163.4697 134.2171
3 0.0010 0.0455 0.0451 262.9287 77.9663 158.7633 129.3240
4 0.0010 0.0450 0.0446 258.5224 78.8347 155.1028 125.5446
5 0.0010 0.0444 0.0464 260.4237 76.2741 165.9544 122.6475
6 0.0010 0.0434 0.0460 256.9943 73.1171 166.4856 120.4759
7 0.0010 0.0427 0.0450 242.2475 62.9558 159.7364 113.8730
8 0.0010 0.0421 0.0448 242.2933 60.2010 164.9709 112.0135
9 0.0010 0.0416 0.0490 261.9519 67.2519 180.8174 116.6908
10 0.0010 0.0411 0.0442 232.4256 58.0736 155.6501 107.8155
11 0.0009 0.0407 0.0484 254.3649 64.6168 177.8439 111.1913
12 0.0009 0.0403 0.0441 231.0975 57.9405 155.2385 106.6227
13 0.0009 0.0400 0.0398 210.9901 44.8900 148.0531 95.9204 14 0.0009 0.0398 0.0395 205.5291 43.5988 144.2484 93.1630 15 0.0009 0.0396 0.0394 203.4009 43.5219 142.4611 91.9637 16 0.0009 0.0393 0.0395 206.3032 45.2096 144.8397 93.0317 17 0.0009 0.0383 0.0397 209.1935 47.7696 144.4370 95.3269 18 0.0009 0.0375 0.0398 211.0529 46.9361 145.3159 97.2030 19 0.0009 0.0369 0.0404 221.1050 49.7893 155.6050 100.7931
20 0.0009 0.0344 0.0327 193.0582 54.8386 120.1525 94.1617 21 0.0009 0.0292 0.0327 193.1149 53.3500 120.7511 94.9096 22 0.0009 0.0279 0.0312 175.3335 47.9422 106.9134 86.1120 23 0.0009 0.0230 0.0142 149.7090 48.0322 80.1184 85.5575 24 0.0009 0.0213 0.0227 160.5323 46.6662 86.3672 88.4285 25 0.0009 0.0221 0.0143 147.6140 42.9623 82.9987 84.0962 26 0.0009 0.0227 0.0236 164.3380 39.9997 88.7122 95.9165 27 0.0009 0.0225 0.0244 172.4369 41.9159 91.8664 103.0370
28 0.0009 0.0225 0.0310 171.1435 38.9638 107.5898 86.9228 29 0.0009 0.0224 0.0143 146.6273 43.5635 80.1099 83.7891 30 0.0009 0.0226 0.0168 148.8766 43.1993 81.0752 86.8134 31 0.0008 0.0227 0.0170 150.1387 42.7214 79.7891 89.2603 32 0.0008 0.0218 0.0231 156.8367 29.9924 81.6934 97.3653 33 0.0008 0.0211 0.0226 151.8681 29.1885 80.8528 92.6760 34 0.0008 0.0206 0.0219 143.9837 27.4500 82.3356 83.5701 35 0.0008 0.0203 0.0216 140.2606 28.1733 81.1518 79.9869 36 0.0008 0.0201 0.0301 158.9964 30.5110 109.1320 76.8475 37 0.0008 0.0199 0.0139 138.8802 29.8528 71.7832 89.2766 38 0.0008 0.0197 0.0221 148.8764 27.8246 80.2098 90.3598 39 0.0008 0.0196 0.0309 170.6370 32.5793 114.6069 85.4801 40 0.0008 0.0196 0.0296 150.3904 28.5499 103.9346 71.4003 41 0.0008 0.0196 0.0219 144.8593 25.9714 78.9707 86.9441 42 0.0008 0.0194 0.0308 180.4644 31.0993 112.8474 94.4349 43 0.0008 0.0193 0.0217 142.0589 26.0486 84.0369 80.9228 44 0.0008 0.0195 0.0300 156.7642 30.2370 108.3944 75.1436 45 0.0008 0.0194 0.0206 127.6690 24.5958 76.4555 70.1406 46 0.0008 0.0193 0.0217 142.2271 25.5931 79.7171 84.0380 47 0.0008 0.0194 0.0257 158.3893 28.2093 105.9763 77.8680 48 0.0008 0.0194 0.0217 140.8707 25.8976 81.2526 81.5904 49 0.0008 0.0193 0.0209 131.0522 24.9918 79.2263 72.1493 50 0.0008 0.0183 0.0127 125.4807 27.4693 61.8725 82.0131 51 0.0008 0.0177 0.0218 142.1969 28.1916 79.0689 83.3884 52 0.0008 0.0174 0.0207 130.5126 27.1028 78.6379 71.4917 53 0.0008 0.0171 0.0253 156.0843 29.9857 105.7877 75.4285 54 0.0008 0.0165 0.0210 136.1075 27.8805 83.3995 74.4445 55 0.0008 0.0167 0.0164 115.4923 29.4462 58.3001 68.8440 56 0.0008 0.0165 0.0167 119.2759 31.0320 59.4517 71.7606 57 0.0008 0.0158 0.0170 121.5507 31.3322 59.5971 74.2171 58 0.0008 0.0155 0.0214 140.6961 29.0039 85.2836 78.5667 59 0.0008 0.0151 0.0079 101.9661 30.1241 40.0179 70.8736 60 0.0008 0.0151 0.0081 103.4289 31.4168 42.0932 70.5341 61 0.0008 0.0153 0.0081 103.1720 31.4207 43.5024 69.3844 62 0.0008 0.0153 0.0078 100.0911 30.4982 40.9974 67.8399 63 0.0008 0.0152 0.0073 95.6702 28.7104 39.1084 65.1040 64 0.0008 0.0146 0.0044 93.5790 37.6982 33.4564 63.7392 65 0.0008 0.0131 0.0034 79.8929 35.2382 22.0330 56.2041 66 0.0008 0.0111 0.0031 76.2633 32.6173 19.0152 55.6717 67 0.0008 0.0090 0.0093 104.6676 35.9006 47.0869 61.3780 68 0.0008 0.0080 0.0032 77.1620 33.2458 21.2665 54.8034 69 0.0008 0.0088 0.0033 79.5209 34.4729 23.5216 55.4175 70 0.0008 0.0101 0.0101 116.7386 38.3904 56.3477 68.4537 71 0.0007 0.0115 0.0031 78.2559 32.7831 20.6876 56.9449 72 0.0007 0.0125 0.0046 95.2108 37.5515 30.0925 67.8318 73 0.0007 0.0130 0.0047 94.9346 36.8610 26.2805 70.3543 74 0.0007 0.0133 0.0041 88.0026 36.0700 23.1403 64.5176 75 0.0007 0.0134 0.0044 91.5324 35.5612 24.6690 68.1658 76 0.0007 0.0133 0.0041 88.7736 36.6139 23.6259 64.6716 77 0.0007 0.0133 0.0044 92.2180 38.0062 23.6622 67.7693 78 0.0007 0.0131 0.0049 96.8663 39.4191 25.1775 71.2484 79 0.0007 0.0130 0.0037 84.0867 36.4608 23.0168 59.8105 80 0.0007 0.0072 0.0060 72.9188 26.8589 25.3076 51.9114 81 0.0007 0.0036 0.0019 58.5538 23.4213 15.9028 43.3407 82 0.0007 0.0044 0.0022 62.8228 23.6570 19.0906 46.3630 83 0.0007 0.0019 0.0012 42.5921 11.3873 10.9498 35.8252 84 0.0007 0.0008 0.0011 41.2149 9.5125 10.0529 35.7393 85 0.0007 0.0007 0.0011 40.4079 9.0933 10.0935 34.9503 86 0.0007 0.0007 0.0011 41.2802 8.5381 10.0900 36.2682 87 0.0007 0.0007 0.0012 42.1489 8.1483 10.0461 37.4371 88 0.0007 0.0006 0.0010 38.6478 7.9534 9.4427 33.9101 89 0.0007 0.0006 0.0011 41.2209 8.2422 9.8165 36.4824 90 0.0007 0.0006 0.0011 41.1873 8.4111 10.0270 36.1804 91 0.0007 0.0006 0.0010 39.0602 7.8220 9.3079 34.4811 92 0.0007 0.0006 0.0011 40.5784 7.8068 9.2009 36.2452 93 0.0007 0.0005 0.0010 38.9321 7.1776 9.2859 34.6574 94 0.0007 0.0006 0.0010 39.4523 7.7836 9.1239 35.1218 95 0.0007 0.0006 0.0010 38.7000 7.4251 9.2010 34.3731 96 0.0007 0.0006 0.0012 42.9929 7.7988 10.6990 38.0390 97 0.0007 0.0006 0.0011 41.6369 7.6291 9.7266 37.2737 98 0.0007 0.0005 0.0011 39.8569 7.1305 9.4462 35.6223 99 0.0007 0.0005 0.0011 40.5055 7.3636 9.5328 36.2498 100 0.0007 0.0005 0.0010 38.5658 7.1534 8.7827 34.5549 101 0.0006 0.0005 0.0011 39.5677 7.2415 9.1677 35.4072 102 0.0006 0.0005 0.0010 38.5633 6.9752 9.2956 34.3792 103 0.0006 0.0005 0.0010 38.5446 7.0830 9.5657 34.1788 104 0.0006 0.0005 0.0010 37.2928 6.9657 8.5839 33.3495 105 0.0006 0.0005 0.0010 38.4879 6.8281 9.3537 34.3484 106 0.0006 0.0005 0.0010 39.6547 7.0100 8.9043 35.7226 107 0.0006 0.0005 0.0009 36.7887 6.8230 8.6941 32.8011 108 0.0006 0.0005 0.0009 37.1044 6.9808 8.4966 33.1855 109 0.0006 0.0005 0.0010 38.7377 6.7962 9.7909 34.3386 110 0.0006 0.0005 0.0009 36.9481 6.5004 8.9710 32.9606 111 0.0006 0.0005 0.0010 39.4202 7.1164 8.9327 35.4611 112 0.0006 0.0005 0.0010 39.2304 6.9602 8.9966 35.2835 113 0.0006 0.0005 0.0010 38.6872 6.8947 8.8385 34.7753 114 0.0006 0.0005 0.0009 36.8878 6.4702 8.4029 33.1693 115 0.0006 0.0004 0.0010 37.1083 6.5803 8.8476 33.1509 116 0.0006 0.0005 0.0009 36.9524 6.7076 8.9353 32.8919 117 0.0006 0.0005 0.0010 38.1768 6.8704 8.5666 34.4101 118 0.0006 0.0005 0.0011 40.5675 7.1207 9.3654 36.4728 119 0.0006 0.0004 0.0011 39.5347 6.9575 8.7204 35.7191 120 0.0006 0.0004 0.0010 39.6086 7.1353 9.0722 35.6381 121 0.0006 0.0004 0.0011 40.9208 7.2228 9.9164 36.4892 122 0.0006 0.0004 0.0010 37.9825 6.5831 9.1964 33.9974 123 0.0006 0.0004 0.0010 37.3988 6.5226 8.6978 33.6241 124 0.0006 0.0004 0.0010 38.7751 6.7058 9.3017 34.7021 125 0.0006 0.0004 0.0011 40.5906 6.7928 8.6735 36.9574 126 0.0006 0.0004 0.0010 38.5451 6.6345 8.4700 34.8802 127 0.0006 0.0004 0.0011 40.2900 6.8071 8.7100 36.5894 128 0.0006 0.0004 0.0010 38.0694 6.7358 8.7022 34.2336 129 0.0006 0.0004 0.0011 39.9285 6.7376 8.8134 36.1903 130 0.0006 0.0004 0.0011 39.9668 6.8143 8.7342 36.2042 131 0.0006 0.0004 0.0010 38.1238 6.6234 9.7511 33.7011 132 0.0006 0.0004 0.0010 38.6865 6.8108 9.7565 34.2221 133 0.0006 0.0004 0.0010 38.4818 6.4918 9.4102 34.2990 134 0.0006 0.0005 0.0011 39.9790 7.0938 9.8880 35.4539 135 0.0006 0.0005 0.0010 38.0519 6.7158 9.9445 33.4711 136 0.0006 0.0005 0.0010 37.6644 6.6367 9.3893 33.4225 137 0.0006 0.0006 0.0010 38.0355 6.6477 8.9835 34.0526 138 0.0006 0.0005 0.0010 39.5439 6.6812 9.2225 35.5593 139 0.0006 0.0005 0.0010 37.8412 6.2307 9.5664 33.6834 140 0.0006 0.0005 0.0009 36.3864 6.3857 8.7918 32.4472

kyang-06 commented 10 months ago

This issue is moving in the direction of a duplicate of #5. To avoid confusion and consolidate discussion, I will close this one.