https://github.com/boostcampaitech5/level1_imageclassification-cv-09

level1_imageclassification-cv-09 created by GitHub Classroom

Science Score: 10.0%

This score indicates how likely this project is to be science-related based on various indicators:

○
CITATION.cff file
○
codemeta.json file
○
.zenodo.json file
○
DOI references
✓
Academic publication links
Links to: arxiv.org
○
Academic email domains
○
Institutional organization owner
○
JOSS paper metadata
○
Scientific vocabulary similarity
Low similarity (6.9%) to scientific vocabulary

Last synced: 10 months ago · JSON representation

Repository

level1_imageclassification-cv-09 created by GitHub Classroom

Basic Info

Host: GitHub
Owner: boostcampaitech5
Language: Python
Default Branch: main
Size: 41.8 MB

Statistics

Stars: 0
Watchers: 0
Forks: 0
Open Issues: 0
Releases: 0

Created over 3 years ago · Last pushed about 3 years ago

https://github.com/boostcampaitech5/level1_imageclassification-cv-09/blob/main/

# Image Classication Competition 
![image](https://user-images.githubusercontent.com/72616557/228166051-e8197cb8-0025-485d-becc-cba4a5c257fd.png)

## About
 :    1.    2.  3.    18 class  



 
#ViT32  #ViT16  #ModelSoups  #Relabeling #Oversampling #ContrastiveLearnig  #WeightedAverageEnsemble    #HardVoting  #SoftVoting  #Optuna  #Wandb 

****





  
## Setting Step
### 1.     
```bash
conda env create -f environment.yml
conda activate model_soups
```
### 2.   
- wandb, albumentations     
### 3. pretrained model   
- [Model soups](https://github.com/mlfoundations/model-soups/releases/tag/v0.0.2)  ViT-B/32  
-  72,    40 .
```bash
python main.py --download-models --model-location   
```
****
## Training Step  
### 1. ViT-B/32, ViT-B/16
#### 1-1. Fine Tuning
```bash
python finetune.py --name {} --i { number} --random-seed { }
```
- [Model soups](https://github.com/mlfoundations/model-soups/releases/tag/v0.0.2)  pretrained  18 class vector output  1 linear layer  . 
- ViT-B/16   Model soups pretrained weight   clip   ImageNet pretrained weight  .
- `--model {ViT-B/32 | ViT-B/16}` : base  
- `--name` :   
- `--i` : pretrained model index
- `--random-seed` : random seed
- `--lr`, `--batch-size`, `--epochs`, `--data-location`, `--model-location` : learning rate, batch size, epoch,  ,   

- Tip :     . training.sh      

```bash
bash training.sh
``` 
#### 1-2. Data oversampling

- Age  Old class  train dataset      Old class data  Over sampling.
- `--old-aug True` : Old class 1  over sampling

#### 1-3. Loss Function 

- Interclass  , Intraclass   Contrastive Learning .
- `--loss-fn` : ContrastiveLoss or CrossEntropyLoss, default CrossEntropyLoss


### 2. Model Soups
- [Model soups](https://github.com/mlfoundations/model-soups/releases/tag/v0.0.2)      pretrained         . 
-    .
1.  pretrained model Test  Accuracy .
2. Accuracy   .
3.    weight average   .
4.          Accuracy   , 3, 4 .   average  3, 4 .
5.  Accuracy     . 
#### 2-1. Fine Tuning

- [Model soups](https://github.com/mlfoundations/model-soups/releases/tag/v0.0.2)  pretrained model ViT-B/32 .
- 1  Fine tuning .


#### 2-2. Individual Evaluation  
```bash
python main.py --eval-individual-models --name {} --model-num { } --random-seed { }
```
- finetune    accuracy  . 
- `--name` :  
- `--model-num` : Evaludation  
- `--random-seed` :   
- `--val-ratio`, `--epoch`, `--data-location`, `--model-locatoin` : validation dataset , epoch,  ,   

-   logs     accuracy  jsonl  . 

#### 2-3. Greedy Soup
```bash
python main.py --greedy-soup --name {} --model-num { } --random-seed { }
```  
- individual Evaluation    accuracy  .         greedy (averaging)        .
- `--name` :  
- `--model-num` : Evaludation  
- `--random-seed` :   
- `--val-ratio`, `--epoch`, `--data-location`, `--model-locatoin` : validation dataset , epoch,  ,   
-   model     .
- log    GREEDY_SOUP_LOG_FILE   .   averaging   .

****
## Inference Step
### 1.    w/ Validation dataset
```bash
python validation.py --model-name {.pt }
```
- Validation set class    .
-    ,  random seed        .
- `--model-name` : evaluation , 
- `--i` : pretrained model index
- `--random-seed` :  


#### 1-1. Weighted Average Ensemble  
- Age class    Age    ,   class(18)    weighted sum .
- `--weighted-ensemble` : Age class  , Default None
```
python finetune_age.py --name {} --i { number} --random-seed { }
```
- finetune_age.py Age class .
- `--name`, `--i`, `--random-seed` finetune.py  
#### 1-2. Soft voting (Ensemble)  
- 2    class  minmax scaling   . 
- `--soft-voting` : soft voting , Default None 
#### 1-3. Hard voting (Ensemble)
-    csv    Hard voting  Ensemble .
- hard_voting.ipynb  ,   csv  hard voting   . 

   .   




### 2. Test w/ Test dataset
```bash
python inference.py --model-name {.pt }
```
-   (.pt)  Test data  .  
- `--model-name` : inference 
- `--weighted-ensemble`, `--soft-voting` : Weighted average ensemble  , Soft Voting   

-   csv  output  . 


****
## Additional Step
### 1. Dataset Relabeling  
![image](https://user-images.githubusercontent.com/113486402/233954582-70a43065-7586-483e-abf5-707e744eebb3.png)  

-    id   relabel_dict   Relabeling .


### 2. Hyperparameter Tuning  
```bash
python optuna_script.py
```
- Optuna  Hyper paramter tuning .
- optuna_script.py  hyper parameter tuning       .  

****
## Result
- Private score 3rd / F1 score - 0.7613 / Accuracy - 81.3175
- Public score 6th / F1 score - 0.7653 / Accuracy - 81.3968
![  2023-04-26 022440](https://user-images.githubusercontent.com/33598545/234355466-63a4c6c0-1b86-4039-a327-15bcf7758db1.png)


****


## Contributors

| |                                                  | |                                                  ||
|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------:|:----------------------------------------------------------------------------------------------------------------------------------------------------------------------:|
| [](https://github.com/june95) 
 | [](https://github.com/Hyunmin-H) 
 | [](https://github.com/hyuns66) 
 | [](https://github.com/jibeomkim7) 
 |[](https://github.com/jennifer060697) 
 |

****
## Reference


Model soups : [Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time](https://arxiv.org/abs/2203.05482).

ViT : https://github.com/google-research/vision_transformer

ContrastiveLoss : https://github.com/KevinMusgrave/pytorch-metric-learning

Optuna : https://optuna.org/

albumentations : https://albumentations.ai/

PyTorch : https://pytorch.org/

Wandb : https://wandb.ai/site