andrei-saceleanu
commited on
Commit
•
4837300
1
Parent(s):
799a68f
Update README.md
Browse files
README.md
CHANGED
@@ -1,47 +1,43 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
-
tags:
|
4 |
-
- generated_from_keras_callback
|
5 |
model-index:
|
6 |
- name: vit-base-vocalsound-logmel
|
7 |
results: []
|
8 |
---
|
9 |
|
10 |
-
<!-- This model card has been generated automatically according to the information Keras had access to. You should
|
11 |
-
probably proofread and complete it, then remove this comment. -->
|
12 |
-
|
13 |
# vit-base-vocalsound-logmel
|
14 |
|
15 |
-
This model is a fine-tuned version of [google/vit-base-patch16-224](https://huggingface.co/google/vit-base-patch16-224) on
|
16 |
It achieves the following results on the evaluation set:
|
17 |
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
## Intended uses & limitations
|
24 |
-
|
25 |
-
More information needed
|
26 |
|
27 |
## Training and evaluation data
|
28 |
|
29 |
-
|
30 |
|
31 |
-
|
32 |
|
33 |
### Training hyperparameters
|
34 |
|
35 |
The following hyperparameters were used during training:
|
36 |
-
- optimizer:
|
|
|
|
|
|
|
37 |
- training_precision: float32
|
38 |
|
39 |
-
###
|
40 |
-
|
41 |
|
|
|
|
|
42 |
|
43 |
### Framework versions
|
44 |
|
45 |
- Transformers 4.27.4
|
46 |
- TensorFlow 2.12.0
|
47 |
-
- Tokenizers 0.13.3
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
3 |
model-index:
|
4 |
- name: vit-base-vocalsound-logmel
|
5 |
results: []
|
6 |
---
|
7 |
|
|
|
|
|
|
|
8 |
# vit-base-vocalsound-logmel
|
9 |
|
10 |
+
This model is a fine-tuned version of [google/vit-base-patch16-224](https://huggingface.co/google/vit-base-patch16-224) on [VocalSound](https://github.com/YuanGongND/vocalsound) dataset.
|
11 |
It achieves the following results on the evaluation set:
|
12 |
|
13 |
+
- accuracy: 88.8
|
14 |
+
- precision (micro): 91.3
|
15 |
+
- recall (micro): 87.1
|
16 |
+
- f1 score (micro): 89.1
|
17 |
+
- f1 score (macro): 89.1
|
|
|
|
|
|
|
18 |
|
19 |
## Training and evaluation data
|
20 |
|
21 |
+
Training: VocalSound training split (#samples = 15570)
|
22 |
|
23 |
+
Evaluation: VocalSound test split(#samples = 3594)
|
24 |
|
25 |
### Training hyperparameters
|
26 |
|
27 |
The following hyperparameters were used during training:
|
28 |
+
- optimizer: AdamW
|
29 |
+
- weight_decay: 0
|
30 |
+
- learning_rate: 5e-5
|
31 |
+
- batch_size: 32
|
32 |
- training_precision: float32
|
33 |
|
34 |
+
### Preprocessing
|
|
|
35 |
|
36 |
+
Differently from [vit-base-vocalsound](https://huggingface.co/andrei-saceleanu/vit-base-vocalsound), the log-melspectrogram is used(log was applied as an addition) and the preprocessor normalization
|
37 |
+
step uses VocalSound statistics(i.e. mean and std) instead of the default IMAGENET ones.
|
38 |
|
39 |
### Framework versions
|
40 |
|
41 |
- Transformers 4.27.4
|
42 |
- TensorFlow 2.12.0
|
43 |
+
- Tokenizers 0.13.3
|