Update README.md
Browse files
README.md
CHANGED
@@ -12,11 +12,13 @@ Our base model is SOTA-comparable on multiple MIR tasks even under probing setti
|
|
12 |
Larger models trained with more data are on the way~
|
13 |
|
14 |
# Model Architecture
|
|
|
15 |
Music2Vec Framework. During pre-training, the student model aims to
|
16 |
reconstruct the masked music audio by taking the contextualized representations provided by the teacher model as prediction targets.
|
17 |
![Model Architecture](music2vec.png)
|
18 |
|
19 |
# Performance Comparison
|
|
|
20 |
With 95M parameters and relatively small training data (1k hr), our base Music2Vec representation achieves comparable performance to the SOTA Jukebox-5B representation.
|
21 |
Note that our base model size is **<2%** of Jukebox-5B.
|
22 |
![Performance Comparison](music2vec_performance.png)
|
|
|
12 |
Larger models trained with more data are on the way~
|
13 |
|
14 |
# Model Architecture
|
15 |
+
|
16 |
Music2Vec Framework. During pre-training, the student model aims to
|
17 |
reconstruct the masked music audio by taking the contextualized representations provided by the teacher model as prediction targets.
|
18 |
![Model Architecture](music2vec.png)
|
19 |
|
20 |
# Performance Comparison
|
21 |
+
|
22 |
With 95M parameters and relatively small training data (1k hr), our base Music2Vec representation achieves comparable performance to the SOTA Jukebox-5B representation.
|
23 |
Note that our base model size is **<2%** of Jukebox-5B.
|
24 |
![Performance Comparison](music2vec_performance.png)
|