metadata

tags:
  - image-classification
  - sign-language
  - indonesian
library_name: pytorch
base_model: google/efficientnet-b3
language:
  - id
metrics:
  - accuracy
pipeline_tag: image-classification

EfficientNet-B3 BISINDO Indonesian Sign Language Classifier

Trained on 9169 images with 26 classes (A-Z BISINDO (Bahasa Isyarat Indonesia) sign language).

Usage

model = models.efficientnet_b3(pretrained=False)
model.classifier = nn.Sequential(
    nn.Dropout(p=0.4),
    nn.Linear(1536, 512),
    nn.SiLU(),
    nn.Dropout(p=0.3),
    nn.Linear(512, 26)
)
model.load_state_dict(torch.load('efficientnet_bisindo_sign_language.pth'))

Training Result

Epoch	Train Loss	Train Acc	Val Loss	Val Acc
1	1.5356	0.7389	1.0504	0.9106
2	1.0970	0.8798	0.9385	0.9248
3	1.0251	0.8945	0.8828	0.9438
4	0.9710	0.9177	0.8689	0.9477
5	0.9382	0.9256	0.8370	0.9597
6	0.9161	0.9365	0.8222	0.9656
7	0.9052	0.9393	0.8133	0.9651
8	0.8904	0.9464	0.8094	0.9635
9	0.8756	0.9460	0.8023	0.9738
10	0.8656	0.9521	0.7939	0.9744
11	0.8609	0.9560	0.7836	0.9749
12	0.8544	0.9545	0.7876	0.9755
13	0.8396	0.9654	0.7929	0.9755
14	0.8415	0.9628	0.7776	0.9798
15	0.8333	0.9654	0.7827	0.9722
16	0.8286	0.9676	0.7695	0.9815
17	0.8235	0.9695	0.7761	0.9853
18	0.8155	0.9715	0.7714	0.9820
19	0.8151	0.9748	0.7589	0.9826
20	0.8139	0.9730	0.7756	0.9776
21	0.8018	0.9791	0.7613	0.9847
22	0.7942	0.9789	0.7475	0.9869
23	0.7859	0.9823	0.7458	0.9864
24	0.7871	0.9802	0.7538	0.9869
25	0.7880	0.9810	0.7525	0.9880