[問題] torch的backward是怎麼管理的?
考慮一個類似STN(Spatial Transform Network)搭配一個用上CNN層的Classfier的架構
只是STN這次是靠不同於Classifier,獨立的loss function去評價並產生loss餵給STN自己
這樣的話怎麼把兩個backward分開進行?
是靠各自弄一個optimizer來管理嗎?
--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 42.77.107.182 (臺灣)
※ 文章網址: https://www.ptt.cc/bbs/DataScience/M.1712816645.A.764.html
推
04/12 19:45,
8月前
, 1F
04/12 19:45, 1F
→
04/12 19:45,
8月前
, 2F
04/12 19:45, 2F
→
04/12 19:45,
8月前
, 3F
04/12 19:45, 3F
→
04/13 00:43,
8月前
, 4F
04/13 00:43, 4F
→
04/13 00:43,
8月前
, 5F
04/13 00:43, 5F
→
04/13 00:43,
8月前
, 6F
04/13 00:43, 6F
→
04/13 00:43,
8月前
, 7F
04/13 00:43, 7F
→
04/13 00:58,
8月前
, 8F
04/13 00:58, 8F
推
04/13 06:04,
8月前
, 9F
04/13 06:04, 9F
→
04/13 06:04,
8月前
, 10F
04/13 06:04, 10F
→
04/13 06:04,
8月前
, 11F
04/13 06:04, 11F
→
04/13 13:59,
8月前
, 12F
04/13 13:59, 12F
→
04/13 13:59,
8月前
, 13F
04/13 13:59, 13F
→
04/13 13:59,
8月前
, 14F
04/13 13:59, 14F
→
04/13 13:59,
8月前
, 15F
04/13 13:59, 15F
→
04/13 13:59,
8月前
, 16F
04/13 13:59, 16F
→
04/13 14:00,
8月前
, 17F
04/13 14:00, 17F
→
04/13 14:00,
8月前
, 18F
04/13 14:00, 18F
→
04/13 14:02,
8月前
, 19F
04/13 14:02, 19F
→
04/13 14:02,
8月前
, 20F
04/13 14:02, 20F
→
04/13 14:02,
8月前
, 21F
04/13 14:02, 21F
→
04/13 14:03,
8月前
, 22F
04/13 14:03, 22F
→
04/13 14:03,
8月前
, 23F
04/13 14:03, 23F
→
04/13 14:08,
8月前
, 24F
04/13 14:08, 24F
→
04/13 14:13,
8月前
, 25F
04/13 14:13, 25F
推
04/13 15:10,
8月前
, 26F
04/13 15:10, 26F
→
04/13 15:10,
8月前
, 27F
04/13 15:10, 27F
→
04/13 17:05,
8月前
, 28F
04/13 17:05, 28F
→
04/26 03:31,
8月前
, 29F
04/26 03:31, 29F
DataScience 近期熱門文章
PTT數位生活區 即時熱門文章