Networks Accuracy trained

Paper	Dataset	Network	Bit(A/W/G)	Paper report	My Accuracy	Comment
-	imagenet	v1	32/32/32	-	69.204	cbas-0
-	imagenet	v1	32/32/32	-	69.800	cbas-1
-	imagenet	v1	32/32/32	-	69.178	cbas-2
-	imagenet	v1	32/32/32	-	65.568	cbas-3
-	imagenet	v1	32/32/32	-	72.446	cbas-4
-	imagenet	v1	32/32/32	-	73.164	cbas-5
-	imagenet	v1	32/32/32	-	55.780	cbas-5, group-norm=8
-	imagenet	v1	32/32/32	-	58.246	cbas-5, group-norm=16
-	imagenet	v1	32/32/32	-	70.310	bacs-1
-	imagenet	v1	32/32/32	-	67.200	bacs-3
-	imagenet	v1	32/32/32	-	71.652	bacs-4
-	imagenet	v2	32/32/32	-	70.416	cbas-1
-	imagenet	v2	32/32/32	-	72.288	cbas-4
-	imagenet	v2	32/32/32	-	71.800	bacs-5
	imagenet	-	-	-	-
dorefa	imagenet	v1	4/4/32	-	51.940	cbas-adam-0
lqnet	imagenet	v1	2/2/32	-	68.094	cbas-sgd-0 dpo wg1fg1
lqnet	imagenet	v1	2/2/32	-	67.832	cbas-sgd-0 dpo wg8fg8
lqnet	imagenet	v1	2/2/32	-	59.648	cbas-sgd-1 dpo wg1fg1 continue with pt all para same lr
lqnet	imagenet	v1	2/2/32	-	60.670	cbas-sgd-0 dpo wg1fg1 continue with pt custom lr
lqnet	imagenet	v1	2/2/32	-	60.360	cbas-sgd-0 directly wg1fg1
lqnet	imagenet	v1	2/2/32	-	66.136	cbas-sgd-0 pto fg8
lqnet	imagenet	v1	2/2/32	-	66.200	cbas-sgd-0 pto fg1
lqnet	imagenet	v1	2/2/32	-	65.086	cbas-sgd-1 pto fg1
	cifar100	-	-	-	-
-	cifar100	v1	16/16/16	-	71.040	cbas-0
-	cifar100	v1	32/32/32	-	70.900	cbas-0
-	cifar100	v1	32/32/32	-	72.070	cbas-1
-	cifar100	v1	32/32/32	-	72.570	cbas-2

TAGS: experiments training-trick quantization classification