@@ -32,28 +32,28 @@ sh job_dynamic.sh #You can change model at line 5 and threshold at line 40
32
32
33
33
## Results
34
34
35
- | ` dataset ` | ` model ` | ` mthresh ` | ` mode ` | Accuracy | FLOPS Reduction (%) |
36
- | ---------- | ------------- | ------------------------- | ----------- | -------- | ------------------- |
37
- | ` cifar10 ` | ` vgg16-bn ` | Baseline Accuracy: 93.82% |
38
- | | | 0.92 | ` joint ` | 93.55% | 65% |
39
- | | | 0.92 | ` decoupled ` | 93.73% | 56% |
40
- | | | 0.85 | ` decoupled ` | 93.19% | 73% |
41
- | | | 0.88 | ` joint ` | 92.65% | 74% |
42
- | | ` resnet56 ` | Baseline Accuracy: 93.66% |
43
- | | | 0.80 | ` decoupled ` | 92.63% | 66% |
44
- | | | 0.88 | ` joint ` | 92.28% | 54% |
45
- | | ` mobilenetv1 ` | Baseline Accuracy: 90.89% |
46
- | | | 1.00 | ` decoupled ` | 91.06% | 78% |
47
- | | | 1.00 | ` joint ` | 91.21% | 78% |
48
- | ` imagenet ` | ` resnet34 ` | Baseline Accuracy: 73.30% |
49
- | | | 0.97 | ` decoupled ` | 73.25% | 25.86% |
50
- | | | 0.95 | ` decoupled ` | 72.79% | 37.77% |
51
- | | | 0.93 | ` decoupled ` | 72.17% | 47.42% |
52
- | | | 0.92 | ` decoupled ` | 71.71% | 52.24% |
53
- | | ` resnet18 ` | Baseline Accuracy: 69.76% |
54
- | | | 0.91 | ` decoupled ` | 67.49% | 51.56% |
55
- | | ` mobilenetv1 ` | Baseline Accuracy: 69.57% |
56
- | | | 1.00 | ` decoupled ` | 69.66% | 41.07% |
35
+ | ` dataset ` | ` model ` | ` mthresh ` | ` mode ` | Accuracy | FLOPS Reduction (%) |
36
+ | ---------- | ------------- | ----------- | ----------- | -------- | ------------------- |
37
+ | ` cifar10 ` | ` vgg16-bn ` | | | 93.82% | Baseline |
38
+ | | | 0.92 | ` joint ` | 93.55% | 65% |
39
+ | | | 0.92 | ` decoupled ` | 93.73% | 56% |
40
+ | | | 0.85 | ` decoupled ` | 93.19% | 73% |
41
+ | | | 0.88 | ` joint ` | 92.65% | 74% |
42
+ | | ` resnet56 ` | | | 93.66% | Baseline |
43
+ | | | 0.80 | ` decoupled ` | 92.63% | 66% |
44
+ | | | 0.88 | ` joint ` | 92.28% | 54% |
45
+ | | ` mobilenetv1 ` | | | 90.89% | Baseline |
46
+ | | | 1.00 | ` decoupled ` | 91.06% | 78% |
47
+ | | | 1.00 | ` joint ` | 91.21% | 78% |
48
+ | ` imagenet ` | ` resnet34 ` | | | 73.30% | Baseline |
49
+ | | | 0.97 | ` decoupled ` | 73.25% | 25.86% |
50
+ | | | 0.95 | ` decoupled ` | 72.79% | 37.77% |
51
+ | | | 0.93 | ` decoupled ` | 72.17% | 47.42% |
52
+ | | | 0.92 | ` decoupled ` | 71.71% | 52.24% |
53
+ | | ` resnet18 ` | | | 69.76% | Baseline |
54
+ | | | 0.91 | ` decoupled ` | 67.49% | 51.56% |
55
+ | | ` mobilenetv1 ` | | | 69.57% | Baseline |
56
+ | | | 1.00 | ` decoupled ` | 69.66% | 41.07% |
57
57
58
58
59
59
## Citation
0 commit comments