search

amdegroot / ssd.pytorch

A PyTorch Implementation of Single Shot MultiBox Detector
MIT License
5.14k stars 1.75k forks source link

train VOC dataset #386

Open wudi00 opened 5 years ago

wudi00 commented 5 years ago

I train VOC dataset wit SSD 300, but loss is too high, why? I have modified the batchsize and learning rate. iter 1680 || Loss: 1885.2950 || timer: 0.1291 sec. iter 1690 || Loss: 2257.7227 || timer: 0.1306 sec. iter 1700 || Loss: 1809.1299 || timer: 0.1481 sec. iter 1710 || Loss: 1642.0751 || timer: 0.1394 sec. iter 1720 || Loss: 1866.5345 || timer: 0.1319 sec. iter 1730 || Loss: 2454.3920 || timer: 0.1527 sec. iter 1740 || Loss: 2005.4082 || timer: 0.1345 sec. iter 1750 || Loss: 1864.5095 || timer: 0.1418 sec. iter 1760 || Loss: 2501.2754 || timer: 0.1366 sec. iter 1770 || Loss: 1213.5183 || timer: 0.1323 sec. iter 1780 || Loss: 1319.9740 || timer: 0.1283 sec. iter 1790 || Loss: 1813.3102 || timer: 0.1406 sec. iter 1800 || Loss: 1719.5035 || timer: 0.1365 sec. iter 1810 || Loss: 2439.4969 || timer: 0.1342 sec. iter 1820 || Loss: 1598.4881 || timer: 0.1377 sec. iter 1830 || Loss: 1470.9045 || timer: 0.1405 sec. iter 1840 || Loss: 1889.6567 || timer: 0.1430 sec. iter 1850 || Loss: 1610.1531 || timer: 0.1399 sec.

yyx1107 commented 5 years ago

没遇到诶，我的是降的

jaychou790118 commented 5 years ago

maybe trouble is your datasets.

haozhihang123 commented 4 years ago

iter 0 || Loss: 25.6081 || timer: 0.0526 sec. iter 10 || Loss: 39.6798 || timer: 0.0514 sec. iter 20 || Loss: 148.1792 || timer: 0.0522 sec. iter 30 || Loss: nan || timer: 0.0515 sec. iter 40 || Loss: nan || timer: 0.0526 sec. iter 50 || Loss: nan || timer: 0.0508 sec. iter 60 || Loss: nan || timer: 0.0511 sec. iter 70 || Loss: nan || timer: 0.0515 sec. iter 80 || Loss: nan || timer: 0.0531 sec. iter 90 || Loss: nan || timer: 0.0527 sec. My result is like this. Has anyone ever been in this situation?

Royee-Lyu commented 3 years ago

iter 0 || Loss: 25.6081 || timer: 0.0526 sec. iter 10 || Loss: 39.6798 || timer: 0.0514 sec. iter 20 || Loss: 148.1792 || timer: 0.0522 sec. iter 30 || Loss: nan || timer: 0.0515 sec. iter 40 || Loss: nan || timer: 0.0526 sec. iter 50 || Loss: nan || timer: 0.0508 sec. iter 60 || Loss: nan || timer: 0.0511 sec. iter 70 || Loss: nan || timer: 0.0515 sec. iter 80 || Loss: nan || timer: 0.0531 sec. iter 90 || Loss: nan || timer: 0.0527 sec. My result is like this. Has anyone ever been in this situation?

Decrease your learning rate