Example api

[1]:

from adelecv.api.logs import enable_logs
from adelecv.api.data.segmentations import SegmentationDataset
from logging import StreamHandler

To enable logs, there is a function python enable logs You can pass any necessary logging hundler

[2]:

enable_logs(StreamHandler())

[3]:

import torch
torch.cuda.is_available()

[3]:

True

[4]:

from adelecv.api.config import Settings
from pathlib import Path

[5]:

# you can change the path to save logs and weights
# Settings.update_tmp_path(Path('your_path'))

[6]:

from adelecv.api.config import Settings

[7]:

Settings.TMP_PATH

[7]:

WindowsPath('tmp/9fa1cc64111e44bb9dfe9650797dcf25')

  1. You need to create a dataset and pass the necessary parameters to it

[8]:

from adelecv.api.data.segmentations.types import ImageMask

[9]:

segm_dataset = SegmentationDataset(
    r'F:\dataset\ph2',  # path to dataset
    ImageMask,          # dataset type
    (256, 256),         # img size HxW
    (0.7, 0.2, 0.1),    # train val test split
    16,                 # batch size
)

 100% |█████████████████| 200/200 [1.3s elapsed, 0s remaining, 152.9 samples/s]

INFO - Split dataset train size: 140, valid size: 40, test size: 20
INFO - Creating a dataset
DEBUG - Dataset created with params, dataset dir: F:\dataset\ph2, classes: {0: 'background', 1: 'target'}, batch size: 16

To get possible values for hyperparameters, there are functions presented below

[10]:

from adelecv.api.models.segmentations import  get_encoders, get_pretrained_weights, get_models, get_torch_optimizers, get_losses, get_optimize_scores
from adelecv.api.optimize.segmentations import get_hp_optimizers
# создать отдельный модуль для получения возможных параметров

[11]:

print('models: ', get_models())
print('encoders: ', get_encoders())
print('pretrained_weights', get_pretrained_weights())
print('optimizers', get_torch_optimizers())
print('losses', get_losses())
print('hp optimizers', get_hp_optimizers())
print('optimize scores', get_optimize_scores())

models:  ['Unet', 'UnetPlusPlus', 'MAnet', 'Linknet', 'FPN', 'PSPNet', 'DeepLabV3', 'DeepLabV3Plus', 'PAN']
encoders:  ['resnet18', 'resnet34', 'resnet50', 'resnet101', 'resnet152', 'resnext50_32x4d', 'resnext101_32x4d', 'resnext101_32x8d', 'resnext101_32x16d', 'resnext101_32x32d', 'resnext101_32x48d', 'dpn68', 'dpn68b', 'dpn92', 'dpn98', 'dpn107', 'dpn131', 'vgg11', 'vgg11_bn', 'vgg13', 'vgg13_bn', 'vgg16', 'vgg16_bn', 'vgg19', 'vgg19_bn', 'senet154', 'se_resnet50', 'se_resnet101', 'se_resnet152', 'se_resnext50_32x4d', 'se_resnext101_32x4d', 'densenet121', 'densenet169', 'densenet201', 'densenet161', 'inceptionresnetv2', 'inceptionv4', 'efficientnet-b0', 'efficientnet-b1', 'efficientnet-b2', 'efficientnet-b3', 'efficientnet-b4', 'efficientnet-b5', 'efficientnet-b6', 'efficientnet-b7', 'mobilenet_v2', 'xception', 'timm-efficientnet-b0', 'timm-efficientnet-b1', 'timm-efficientnet-b2', 'timm-efficientnet-b3', 'timm-efficientnet-b4', 'timm-efficientnet-b5', 'timm-efficientnet-b6', 'timm-efficientnet-b7', 'timm-efficientnet-b8', 'timm-efficientnet-l2', 'timm-tf_efficientnet_lite0', 'timm-tf_efficientnet_lite1', 'timm-tf_efficientnet_lite2', 'timm-tf_efficientnet_lite3', 'timm-tf_efficientnet_lite4', 'timm-resnest14d', 'timm-resnest26d', 'timm-resnest50d', 'timm-resnest101e', 'timm-resnest200e', 'timm-resnest269e', 'timm-resnest50d_4s2x40d', 'timm-resnest50d_1s4x24d', 'timm-res2net50_26w_4s', 'timm-res2net101_26w_4s', 'timm-res2net50_26w_6s', 'timm-res2net50_26w_8s', 'timm-res2net50_48w_2s', 'timm-res2net50_14w_8s', 'timm-res2next50', 'timm-regnetx_002', 'timm-regnetx_004', 'timm-regnetx_006', 'timm-regnetx_008', 'timm-regnetx_016', 'timm-regnetx_032', 'timm-regnetx_040', 'timm-regnetx_064', 'timm-regnetx_080', 'timm-regnetx_120', 'timm-regnetx_160', 'timm-regnetx_320', 'timm-regnety_002', 'timm-regnety_004', 'timm-regnety_006', 'timm-regnety_008', 'timm-regnety_016', 'timm-regnety_032', 'timm-regnety_040', 'timm-regnety_064', 'timm-regnety_080', 'timm-regnety_120', 'timm-regnety_160', 'timm-regnety_320', 'timm-skresnet18', 'timm-skresnet34', 'timm-skresnext50_32x4d', 'timm-mobilenetv3_large_075', 'timm-mobilenetv3_large_100', 'timm-mobilenetv3_large_minimal_100', 'timm-mobilenetv3_small_075', 'timm-mobilenetv3_small_100', 'timm-mobilenetv3_small_minimal_100', 'timm-gernet_s', 'timm-gernet_m', 'timm-gernet_l', 'mit_b0', 'mit_b1', 'mit_b2', 'mit_b3', 'mit_b4', 'mit_b5', 'mobileone_s0', 'mobileone_s1', 'mobileone_s2', 'mobileone_s3', 'mobileone_s4']
pretrained_weights ['imagenet', 'None']
optimizers ['AdamW', 'Adadelta', 'Adam', 'SGD', 'RAdam', 'NAdam', 'RMSprop', 'Adagrad']
losses ['JaccardLoss', 'DiceLoss', 'FocalLoss', 'LovaszLoss', 'SoftBCEWithLogitsLoss', 'SoftCrossEntropyLoss', 'TverskyLoss', 'MCCLoss']
hp optimizers ['RandomSampler', 'GridSampler', 'TPESampler', 'CmaEsSampler', 'NSGAIISampler', 'QMCSampler', 'MOTPESampler']
optimize scores ['fbeta_score', 'f1_score', 'iou_score', 'accuracy', 'positive_predictive_value', 'sensitivity', 'loss']

  1. It is necessary to create a set of hyperparameters, according to which the optimal ones will be searched.

[12]:

from adelecv.api.optimize.segmentations import HyperParamsSegmentation
from adelecv.api.optimize.segmentations import HPOptimizer

[13]:

params = HyperParamsSegmentation(
    strategy='TPESampler',
    architectures=['UnetPlusPlus'],
    encoders=['mobilenet_v2'],
    pretrained_weights=['imagenet'],
    loss_fns=['JaccardLoss'],
    optimizers=['AdamW', 'RMSprop'],
    epoch_range=(5, 5),
    lr_range=(0.001, 0.003),
    optimize_score='loss',
)

  1. You need to create an optimizer object with the transfer of the dataset and hyperparameters to it. You also need to choose an optimization algorithm and set additional training parameters

[15]:

hp_optimizer = HPOptimizer(
    hyper_params=params,
    num_trials=3,
    device='GPU',
    dataset=segm_dataset
)

DEBUG - Create hp optimizer with params: strategy: TPESampler num_trials: 1 num_classes: 2 device: cuda hps: {'strategy': 'TPESampler', 'architectures': ['UnetPlusPlus'], 'encoders': ['mobilenet_v2'], 'pretrained_weights': ['imagenet'], 'loss_fns': ['JaccardLoss'], 'optimizers': ['AdamW', 'RMSprop'], 'epoch_range': (5, 5), 'lr_range': (0.001, 0.003), 'optimize_score': 'loss'}
DEBUG - Create hp optimizer with params: strategy: TPESampler num_trials: 3 num_classes: 2 device: cuda hps: {'strategy': 'TPESampler', 'architectures': ['UnetPlusPlus'], 'encoders': ['mobilenet_v2'], 'pretrained_weights': ['imagenet'], 'loss_fns': ['JaccardLoss'], 'optimizers': ['AdamW', 'RMSprop'], 'epoch_range': (5, 5), 'lr_range': (0.001, 0.003), 'optimize_score': 'loss'}

  1. Tensorboard is used to visualize learning

[16]:

from adelecv.api.config import Settings

[17]:

tensorboard_logs_path = Settings.TENSORBOARD_LOGS_PATH.as_posix()

[17]:

tensorboard_logs_path

[17]:

'tmp/cb0cb0afd3f84fb6b0d8299532ad8759/tensorboard'

[18]:

%reload_ext tensorboard

[19]:

tensorboard_logs_path

[19]:

'tmp/cb0cb0afd3f84fb6b0d8299532ad8759/tensorboard'

[22]:

tensorboard --logdir $tensorboard_logs_path

^C

Launching TensorBoard...

  1. To start the training, call the appropriate method.

[17]:

hp_optimizer.optimize()

INFO - Train models started
[I 2023-05-01 17:46:09,162] A new study created in memory with name: no-name-f6be76f6-855f-4d0f-960b-047ecab722b2
DEBUG - Create tensorboard logger, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/tensorboard/499ba36
DEBUG - Dataset updated
INFO - Model 499ba36_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0022375857163229526 trained with test loss 0.14925605058670044
DEBUG - Add predictions for model: 499ba36_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0022375857163229526

 200 [9.4s elapsed, 20.8 samples/s]

DEBUG - Save weights model: 499ba36_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0022375857163229526, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/weights/499ba36.pt
[I 2023-05-01 17:47:01,928] Trial 0 finished with value: 0.11668000618616739 and parameters: {'optimizer': 'AdamW', 'architecture': 'UnetPlusPlus', 'encoders': 'mobilenet_v2', 'pretrained_weight': 'imagenet', 'lr': 0.0022375857163229526, 'loss': 'JaccardLoss', 'num_epoch': 5}. Best is trial 0 with value: 0.11668000618616739.
DEBUG - Create tensorboard logger, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/tensorboard/d81e347
DEBUG - Dataset updated
INFO - Model d81e347_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0020570884365662253 trained with test loss 0.13842931389808655
DEBUG - Add predictions for model: d81e347_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0020570884365662253

 200 [12.1s elapsed, 15.4 samples/s]

DEBUG - Save weights model: d81e347_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0020570884365662253, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/weights/d81e347.pt
[I 2023-05-01 17:47:54,924] Trial 1 finished with value: 0.10880672931671143 and parameters: {'optimizer': 'AdamW', 'architecture': 'UnetPlusPlus', 'encoders': 'mobilenet_v2', 'pretrained_weight': 'imagenet', 'lr': 0.0020570884365662253, 'loss': 'JaccardLoss', 'num_epoch': 5}. Best is trial 1 with value: 0.10880672931671143.
DEBUG - Create tensorboard logger, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/tensorboard/304731e
DEBUG - Dataset updated
INFO - Model 304731e_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0025402917390305577 trained with test loss 0.09902772307395935
DEBUG - Add predictions for model: 304731e_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0025402917390305577

 200 [15.2s elapsed, 11.6 samples/s]

DEBUG - Save weights model: 304731e_UnetPlusPlus_mobilenet_v2_imagenet_AdamW_JaccardLoss_lr=0,0025402917390305577, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/weights/304731e.pt
[I 2023-05-01 17:48:53,994] Trial 2 finished with value: 0.08640851577123006 and parameters: {'optimizer': 'AdamW', 'architecture': 'UnetPlusPlus', 'encoders': 'mobilenet_v2', 'pretrained_weight': 'imagenet', 'lr': 0.0025402917390305577, 'loss': 'JaccardLoss', 'num_epoch': 5}. Best is trial 2 with value: 0.08640851577123006.
INFO - Study statistics:
INFO - Number of finished trials: 3
INFO - Number of pruned trials: 0
INFO - Number of complete trials: 3
INFO - Best trial:
INFO - Value: 0.08640851577123006
INFO - Params:
INFO - optimizer: AdamW
INFO - architecture: UnetPlusPlus
INFO - encoders: mobilenet_v2
INFO - pretrained_weight: imagenet
INFO - lr: 0.0025402917390305577
INFO - loss: JaccardLoss
INFO - num_epoch: 5
INFO - Train models is over

[19]:

hp_optimizer.stats_models

[19]:
_id name architecture encoder pretrained_weight lr optimizer loss_fn num_epoch loss fbeta_score f1_score iou_score accuracy positive_predictive_value sensitivity
0 499ba36 499ba36_UnetPlusPlus_mobilenet_v2_imagenet_Ada... UnetPlusPlus mobilenet_v2 imagenet 0.002238 AdamW JaccardLoss 5 0.149 0.868 0.868 0.810 0.931 0.900 0.901
1 d81e347 d81e347_UnetPlusPlus_mobilenet_v2_imagenet_Ada... UnetPlusPlus mobilenet_v2 imagenet 0.002057 AdamW JaccardLoss 5 0.138 0.876 0.876 0.823 0.934 0.890 0.929
2 304731e 304731e_UnetPlusPlus_mobilenet_v2_imagenet_Ada... UnetPlusPlus mobilenet_v2 imagenet 0.002540 AdamW JaccardLoss 5 0.099 0.928 0.928 0.884 0.952 0.935 0.944

6 (Optional). A special class is provided for exporting weights (creating a zip archive). Mainly used for the web version of the library

[25]:

from adelecv.api.modification_models import ExportWeights

[ ]:

# you can save the selected weights in a zip file (feature for ui)
# ExportWeights().create_zip()

  1. You can visualize the datasets and prediction of each model using the fiftyone.

[26]:

import fiftyone as fo

[32]:

fo.launch_app(segm_dataset.fo_dataset)

[32]:

Dataset:          ImageMask
Media type:       image
Num samples:      200
Selected samples: 0
Selected labels:  0
Session URL:      http://localhost:5151/

  1. To convert weights, you can use the appropriate classes. Below is an example for the onnx format

[22]:

from adelecv.api.modification_models import ConvertWeights

[23]:

cw = ConvertWeights(input_shape=(1, 3, 256, 256))

[24]:

cw.run(id_selected=['304731e', '499ba36'], new_format='onnx')

INFO - Сonvert weights model: 304731e to onnx format
D:\anaconda\envs\adelecv-check\lib\site-packages\segmentation_models_pytorch\base\model.py:16: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  if h % output_stride != 0 or w % output_stride != 0:
D:\anaconda\envs\adelecv-check\lib\site-packages\torch\onnx\symbolic_helper.py:829: UserWarning: You are trying to export the model with onnx:Resize for ONNX opset version 10. This operator might cause results to not match the expected results by PyTorch.
ONNX's Upsample/Resize operator did not match Pytorch's Interpolation until opset 11. Attributes to determine how to transform the input were added in onnx:Resize in opset 11 to support Pytorch's behavior (like coordinate_transformation_mode and nearest_mode).
We recommend using opset 11 and above for models using this operator.
  warnings.warn(

============= Diagnostic Run torch.onnx.export version 2.0.0+cu117 =============
verbose: False, log level: Level.ERROR
======================= 0 NONE 0 NOTE 0 WARNING 0 ERROR ========================


INFO - Create zip with converted weights, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/converted_bdb48d4c20604da9912eb70d68588744.zip
INFO - Сonvert weights model: 499ba36 to onnx format

============= Diagnostic Run torch.onnx.export version 2.0.0+cu117 =============
verbose: False, log level: Level.ERROR
======================= 0 NONE 0 NOTE 0 WARNING 0 ERROR ========================


INFO - Create zip with converted weights, path: tmp/9fa1cc64111e44bb9dfe9650797dcf25/converted_bdb48d4c20604da9912eb70d68588744.zip

[24]:

WindowsPath('tmp/9fa1cc64111e44bb9dfe9650797dcf25/converted_bdb48d4c20604da9912eb70d68588744.zip')

[ ]: