### Current results :

# VQ-VAE performances

This table contains the performance and parameters of various tests on the vq-vae model on various versions :
* **version 1** : raw model prosed on the original paper :
* **version 2** : model finutuned using the Re-Fit method proposed in the article : , to increase the codeBook usage, this have shown very promissing results since we could attaind 100% usage.
* **version 3** : using ExponentialMocingAverage to update the codebook instead of the MSE loss
* **version 4** : using the RQ-VAE model (link to paper) instead of classic VQ-VAE

|Model version | Model Name | Parameters | Dice Score (%) | N° Codes used | Codebook usage (%) |
|------------------|------------------|-----------------------------------|:--------------:|:-------------:|:------------------:|
|||||||
|V.1 : raw architecture |
|| Model 100 | epochs=100, D= 64 , K= 512 , downsample = 4 , beta = 0.25 | 99.58 | 89 | 17.38 |
|| Model 101 | epochs=200, D= 64 , K= 512 , downsample = 4 , beta = 0.25 | 99.52 | 70 | 13.67 |
|| Model 102 | epochs=100, D= 32 , K= 256 , downsample = 4 , beta = 0.25 | 98.86 | 53 | 20.70 |
|| Model 103 | epochs=100, D= 64 , K= 512 , downsample = 4 , beta = 1 | 91.23 | 40 | 7.8 |
|| Model 104 | epochs=100, D= 64 , K= 256 , downsample = 4 , beta = 0.25 | 98.75 | 22 | 8.59 |
|| Model 105 | epochs=100, D= 64 , K= 512 , downsample = 8 , beta = 0.25 | 95.99 | 49 | 9.57 |
|||||||
|V.2 : Re-Fit finutuning |
|| Model 200 | base_model = v1.model100, epochs=100, D= 64 , K= 256 , downsample = 4 | 99.66 | 256 | 100 |
|| Model 201 | base_model = v1.model100, epochs=100, D= 64 , K= 128 , downsample = 4 | 99.67 | 128 | 100 |
|||||||
|V3. : EMA updates
|| Model 300 | epochs=100, D= 64 , K= 512 , downsample = 4 , beta = 0.25 | 99.70 | 512 | 100 |
|| Model 301 | epochs=100, D= 64 , K= 512 , downsample = 8 , beta = 0.25 | 97.78 | 512 | 100 |

### Here are the performances of the RQ-VAE model : (paper : .link)

| Model Name | Parameters | Dice Score (%) |
|------------------|-----------------------------------|:--------------:|
| Model 400 |D= 64 , K= 512 , downsample = 4 ,num_cb = 2, shared_cb = false, |99.68 |
| Model 401 |D= 64 , K= 512 , downsample = 8 ,num_cb = 2, shared_cb = false, |98.70 |
| Model 400 |D= 64 , K= 256 , downsample = 4 ,num_cb = 2, shared_cb = false, |98.79 |
| Model 400 |D= 64 , K= 512 , downsample = 4 ,num_cb = 2, shared_cb = True , |98.46 |
| Model 400 |D= 64 , K= 512 , downsample = 4 ,num_cb = 4, shared_cb = false, |90.67 |