108334db29
FLOPs update
2020-03-30 16:04:08 -07:00
105882b3c6
GFLOPs correction
2020-03-30 15:30:53 -07:00
de52a008a5
default --img-size to 512
2020-03-30 11:46:20 -07:00
f6fc9634ab
mAP updates
2020-03-30 11:37:38 -07:00
eb151a881e
NMS and test batch_size updates
2020-03-29 20:41:32 -07:00
c6b59a0e8a
LR schedule to 0.05 min
2020-03-29 13:29:06 -07:00
9c5e76b93d
EMA implemented by default
2020-03-29 13:14:54 -07:00
dc8e56b9f3
mAP update
2020-03-28 16:03:46 -07:00
ce17c26759
mAP updates
2020-03-27 13:52:07 -07:00
f9d34587da
Merge NMS update
2020-03-27 13:11:24 -07:00
582de735ad
utils.clip_coords doesn't work as expected. ( #961 )
...
* utils.clip_coords doesn't work as expected.
Box coords may be negative or exceed borders.
* Update utils.py
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
2020-03-27 13:09:10 -07:00
4a63b24b09
Merge NMS update
2020-03-26 19:50:29 -07:00
01ee0c5e95
Merge NMS update
2020-03-26 18:41:04 -07:00
dad59220f1
speed and comments update
2020-03-26 18:34:20 -07:00
faab52913c
mAP updates
2020-03-26 16:35:46 -07:00
9568d4562d
mAP updates
2020-03-26 16:22:58 -07:00
5ab13e5aa2
Merge remote-tracking branch 'origin/master'
2020-03-26 16:20:11 -07:00
470371ba59
Test augment update
2020-03-26 16:20:06 -07:00
20b2671de0
cleanup ( #963 )
...
* cleanup
cleanup
* Update train.py
Co-authored-by: Glenn Jocher <glenn.jocher@ultralytics.com>
2020-03-26 14:46:17 -07:00
a4721e90f8
defult batch-size to 16
2020-03-26 14:22:59 -07:00
4a7d9bdba9
mAP increases
2020-03-26 14:14:52 -07:00
a322fc5d4b
Merge NMS update
2020-03-26 12:48:00 -07:00
171b4129b5
Merge NMS update
2020-03-26 12:33:12 -07:00
eac07f9da3
Merge NMS update
2020-03-26 12:20:01 -07:00
94344f5bea
test augmentation comments
2020-03-26 11:34:32 -07:00
f91b1fb13a
merge_batch NMS method
2020-03-26 11:28:46 -07:00
c71ab7d506
augmented testing
2020-03-26 11:25:44 -07:00
23b34f4db8
merge_batch NMS method
2020-03-25 23:29:33 -07:00
aa0c64b5ac
merge_batch NMS method
2020-03-25 23:24:57 -07:00
3265d50f69
speed update
2020-03-19 18:15:09 -07:00
89b6377723
Fuse by default when test.py called directly (faster)
2020-03-19 18:11:08 -07:00
fff45c39a8
cleanup/speedup
2020-03-19 16:41:42 -07:00
1b68fe7fde
cleanup
2020-03-19 16:23:44 -07:00
83c9cfb7de
FocalLoss() and obj loss speed and stability update
2020-03-19 12:30:37 -07:00
20454990ce
FLOPS report
2020-03-19 12:30:07 -07:00
60c8d194cd
FocalLoss() and obj loss speed and stability update
2020-03-19 11:52:52 -07:00
b3adc896f9
focal and obj loss speed/stability
2020-03-16 21:40:57 -07:00
448c4a6e1f
Remove deprecated --arc architecture options, implement --arc default for all cases
2020-03-16 20:46:25 -07:00
77c6c01970
EMA class updates
2020-03-16 17:51:40 -07:00
1a12667ce1
loss function cleanup
2020-03-16 17:31:37 -07:00
f1208f784e
updated run history
2020-03-16 15:36:56 -07:00
2a12a91245
nvcr.io/nvidia/pytorch:20.02-py3
2020-03-16 15:19:58 -07:00
c09fcfc4fe
EMA class updates
2020-03-16 14:18:56 -07:00
c4047000fe
FocalLoss() updated to match TF
2020-03-16 14:03:50 -07:00
07d2f0ad03
test/inference time augmentation
2020-03-15 18:39:54 -07:00
adba66c3a6
EMA class updates
2020-03-14 18:08:48 -07:00
851c9b9883
EMA class updates
2020-03-14 17:34:13 -07:00
d91469a516
EMA class updates
2020-03-14 17:33:29 -07:00
5ebbb2db28
ASFF implementation
2020-03-14 17:04:38 -07:00
ea4c26b32d
BatchNorm2d() to EfficientDet defaults: decay=0.997 eps=1e-4
2020-03-14 16:54:04 -07:00
Glenn Jocher
GoogleWiki
Yonghye Kwon