0. Quick Start (Get our result)

Set a virtual environment and requirements for our result (Recommended)

$ conda create -n segmentation python=3.7.11
$ conda activate segmentation
$ pip install -r requirements

Execute the following command and get our result.
```
$ sh run.sh
```

1. Prepare

1.1. Get the trash dataset.

$ wget https://aistages-prod-server-public.s3.amazonaws.com/app/Competitions/000078/data/data.zip
$ mv data.zip rawdata.zip
$ unzip rawdata.zip
$ rm rawdata.zip

Dataset Copy Right and License: Naver Connect, CC-BY-2.0
For more detail about this dataset, see here.

1.2. Get Some Libraries.

pytorch-toolbelt (for various loss)
```
$ pip install pytorch_toolbelt
```
For more detail, see here.
MADGRAD (An optimizer)
```
$ pip install madgrad
```
For more detail, see here.

1.3. Completed Structure

./
├─rawdata/
|
├─config/
|    ├─fix_seed.py
|    ├─read_config.py
|    ├─wnb.py
|    └─default.yaml
├─data/
|    ├─dataset.py
|    ├─dataloader.py
|    └─augmentation.py
├─submission/
|    └─sample_submission.csv
├─util/
|    ├─eda.py
|    ├─ploting.py
|    ├─tta.py
|    └─utils.py
|
├─train.py
├─inference.py
├─pipeline.py
|
└─run.sh

2. Pipeline

One Step Execution - From train to inference.

$ python pipeline.py --cfg-yaml ./config/default.yaml

Two Step Execution

Train Step

$ python train.py --cfg-yaml ./config/default.yaml

Inference Step

$ python inference.py --cfg-yaml ./config/default.yaml

3. Configurations

3.1. Configuration file usage

You can see whole configurations in the yaml file (./config/default.yaml)
Copy and paste the yaml file and edit it.
Then, you can get the proper result.

3.2. About configurations

Listing the supported frameworks

FRAMEWORKS_AVAILABLE: ["torchvision", "segmentation_models_pytorch"]

Only support 2 frameworks.

Listing the supported models (including encoders and decoders)

MODELS_AVAILABLE:
    torchvision: ["fcn_resnet50", ... , "lraspp_mobilenet_v3_large"]
                  
DECODER_AVAILABLE: ["unet", ... , "pan"] # smp decoder

ENCODER_AVAILABLE: ['resnet18', ... , 'timm-gernet_l'] # smp encoder

smp : segmentation_models_pytorch

Listing the supported criterion

CRITERION_AVAILABLE:
    # available_framework: [avaliable criterions]
    torch.nn: ["CrossEntropy"]
    pytorch_toolbelt: ["BalancedBCEWithLogitsLoss", ... , "WingLoss"]

Listing the supported K-Fold types.
```
KFOLD_TYPE_AVAILABLE: ["KFold", "MultilabelStratifiedKFold"]
```
For K-Fold reference, see here.

For MLSK-Fold reference, see here.

Model Selection

Torchvision

SELECTED:
    # 1. IF you use torchvision model
    FRAMEWORK: "torchvision"
    MODEL: "lraspp_mobilenet_v3_large" # also used for submission save.
    MODEL_CFG:
        pretrained: True

smp

SELECTED:
    # 2. IF you use smp model
    # smp.create_model(**cfg["SELECTED"]["MODEL_CFG"]) 형태로 사용하기 때문에
    # MODEL_CFG 아래는 소문자가 좋습니다. (PRETRAINED -> pretrained)
    FRAMEWORK: "segmentation_models_pytorch"
    MODEL_CFG:
        arch: "fpn"                          # DECODER
        encoder_name: "timm-efficientnet-b6" # ENCODER (https://smp.readthedocs.io/en/latest/encoders.html) 
        encoder_weights: "noisy-student"     # ENCODER 마다 가능한 DATASET 상이 ("imagenet", "advpros", "noisy-student" 등)
        in_channels: 3 # fixed
        classes: 11    # fixed

Criterion Selection

SELECTED:
  # ...

  CRITERION:
    FRAMEWORK: "pytorch_toolbelt"
    USE: "SoftCrossEntropyLoss"
    CFG:

Experiment configurations

seed, epochs, batch size, learning rate, the number of workers, validation period config

EXPERIMENTS:
    SEED: 21
    NUM_EPOCHS: 30
    BATCH_SIZE: 16
    LEARNING_RATE: 1e-4
    NUM_WORKERS: 4
    VAL_EVERY: 5
    
    # ...

K-Fold config

EXPERIMENTS:
    # ...
    
    KFOLD:
        TURN_ON: True
        TYPE: "MultilabelStratifiedKFold"
        NUM_FOLD: 5
    
    # ...

Autocast

EXPERIMENTS:
    # ...
    
    AUTOCAST_TURN_ON: True
    
    # ...

wandb config

EXPERIMENTS:
    # ...
    
    WNB:
        TURN_ON: True
        INIT:
            entity: "ai_tech_level2-cv-18"
            project: "seunghun_T2042"
            name: "fpn_timm-efficientnet-b6" # recommended to change if wnb is turn-on.
     
     # ...

Configure directories for best performance model saving and submission file saving

EXPERIMENTS:
    # ...
    
    SAVED_DIR: 
        BEST_MODEL: "./saved"
        SUBMISSION: "./submission"
        
    # ...

Configure train transforms, which will be compounded by A.OneOf.

EXPERIMENTS:
    # ...
    
    TRAIN_TRANS: # ToTensorV2 는 기본으로 들어가있고 Albumentation 의 augmentation 이용
        GridDistortion: 
            p: 1.0
        RandomGridShuffle:
            p: 1.0
        RandomResizedCrop:
            height: 512
            width: 512
            p: 1.0
        HorizontalFlip:
            p: 1.0
        VerticalFlip:
            p: 1.0
        GridDropout:
            p: 1.0
        ElasticTransform:
            p: 1.0
            
    # ...

TTA config

EXPERIMENTS:
   # ...
    
    TTA:
        TURN_ON: True
        AVAILABLE_LIST: # only support 2 below TTAs.
        VERTICAL_FLIP_TURN_ON: True
        HORIZONTAL_FLIP_TURN_ON: True

Only support vertical flip and horizontal flip.

(Augmentations are equal to reverse of themselves.)

Dataset configurations

DATASET:
    PATH: "./rawdata" # Config dataset root
    ANNS_FILE_NAME: "train_all.json"
    TRAIN_FILE_NAME: "train_all.json"
    VAL_FILE_NAME: "val.json" # not used if you set "KFOLD TURN ON - True".
    TEST_FILE_NAME: "test.json"
    NUM_CLASSES: 11

4. Our Experiments

4.1. Model

Encoder : timm-efficientnet-b7
- Weight : noisy-student
Decoder : FPN
- In channel : 3
- Classes : 11
Fold : KFold, MultilabelStratifiedKFold
- Number of fold : 5
Learning rate : 2e-4
TTA : Horizontal flip

4.2. Loss

SoftCrossEntropyLoss in pytorch_toolbelt

SoftCE > CE > DiceCE > Dice

4.3. Optimizer

MADGRAD provides generalization performance of SGD and fast convergence speed such as Adam.

MADGRAD > Adam

4.4. Learning rate Scheduler

CosineAnnealingWarmRestarts in torch.optim.lr_scheduler

4.5. Scaler

Autocast and GradScaler were used to shorten training time.

4.6. Augmentations

By using light model, we perform quickly various augmentation experiments.

Selected model : LRASPP mobilenetv3 Large in torchvision - for more detail, see here

Hyperparameter(Epochs) Tuning for the LRASPP mobilenet v3 large model.

Epochs mIoU mIoU derivation

6 epochs 0.510 0.0

12 epochs 0.553 +0.042

24 epochs 0.571 +0.061

For fast experiments, we don't try 48 epochs.

Single Augmentation Observation

Augmentation (Fix 24 epochs)	mIoU	mIoU derivation
None	0.571	0.0
Blur	0.572	+0.001
GridDistortion	0.583	+0.012
RandomGridShuffle	0.585	+0.014
GridDropout	0.587	+0.016
ElasticTransform	0.598	+0.027
RandomResizeCrop	0.619	+0.048

A test about compound augmentation by using albumentations.core.composition.OneOf (see here)
- Use 5 augmentations
  - GridDistortion, RandomGridShuffle, GridDropout, ElasticTransform, RandomResizeCrop
- Result
  
  Epoch (Fix augmentation) mIoU mIoU derivation
  
  24 epochs 0.609 +0.038
  
  48 epochs 0.631 +0.060
  
  96 epochs 0.653 +0.082
  
  As epoch increase, mIoU also increase.

4.7. K-Fold Ensemble

4.8. TTA

We tried to use ttach library but, couldn't use it. So, we apply only flip TTA, which is satisfied that augmentation is equal to reverse augmentation.

You can add such augmentation function codes in ./util/tta.py and modify ./config/default.yaml and get_tta_list function in ./util/tta.py.

4.9. Pseudo labeling

We convert the resulting CSV file into COCO-dataset to apply pseudo labelling.

If you want to modify your path, you should change this part in code.

    # config
    cfg = {
        "csv_file_path" : "", # csv file you want to convert
        "test_file_path" : "", # test_json path
        "result_file_path" : "", # json file you want to save result
        "maxWidth" : 256, # test image width
        "maxHeight" : 256, # test image height
    }

You can use this module in ./util/pseudo.py

5. Result

5.1. Leader Board in Competition

	mIoU
Public LB	`0.781`
Private LB	`0.717`

5.2. Images after model inference

6. Participants

Name	Github	Role
김서기 (T2035)	Link	Research(HRNet, MMSeg library), Pseudo Labeling, TTA
김승훈 (T2042)	Link	Find Augmentations, Code Refactoring
배민한 (T2260)	Link	Research(smp library, loss), Model Enhancement, K-Fold
손지아 (T2113)	Link	Research(smp library, loss), Model Enhancement, MLSK-Fold
이상은 (T2157)	Link	Research(HRNet, optimizer, loss), Pseudo Labeling, Augmix
조익수 (T2213)	Link	Research(MMseg library)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

0. Quick Start (Get our result)

1. Prepare

1.1. Get the trash dataset.

1.2. Get Some Libraries.

1.3. Completed Structure

2. Pipeline

3. Configurations

3.1. Configuration file usage

3.2. About configurations

4. Our Experiments

4.1. Model

4.2. Loss

4.3. Optimizer

4.4. Learning rate Scheduler

4.5. Scaler

4.6. Augmentations

4.7. K-Fold Ensemble

4.8. TTA

4.9. Pseudo labeling

5. Result

5.1. Leader Board in Competition

5.2. Images after model inference

6. Participants

About

Releases

Packages

Contributors 5

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 13 Commits
_SUB		_SUB
config		config
data		data
submission		submission
util		util
.gitignore		.gitignore
DATASET_DESCRIPTION.md		DATASET_DESCRIPTION.md
README.md		README.md
class_dict.csv		class_dict.csv
inference.py		inference.py
pipeline.py		pipeline.py
requirements.txt		requirements.txt
run.sh		run.sh
train.py		train.py

Epochs	mIoU	mIoU derivation
6 epochs	0.510	0.0
12 epochs	0.553	+0.042
24 epochs	0.571	+0.061

Epoch (Fix augmentation)	mIoU	mIoU derivation
24 epochs	0.609	+0.038
48 epochs	0.631	+0.060
96 epochs	0.653	+0.082

boostcampaitech2/semantic-segmentation-level2-cv-18

Folders and files

Latest commit

History

Repository files navigation

0. Quick Start (Get our result)

1. Prepare

1.1. Get the trash dataset.

1.2. Get Some Libraries.

1.3. Completed Structure

2. Pipeline

3. Configurations

3.1. Configuration file usage

3.2. About configurations

4. Our Experiments

4.1. Model

4.2. Loss

4.3. Optimizer

4.4. Learning rate Scheduler

4.5. Scaler

4.6. Augmentations

4.7. K-Fold Ensemble

4.8. TTA

4.9. Pseudo labeling

5. Result

5.1. Leader Board in Competition

5.2. Images after model inference

6. Participants

About

Resources

Stars

Watchers

Forks

Releases

Packages 0

Contributors 5

Languages

Packages