Copy deployment files to gradio branch

LICENSE

@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2023 Javi Carnero
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.

README.md

@@ -1,17 +0,0 @@
-# birds-classification
-
-Train model for birds classification and gradio app
-
-Training is done using fastai, deployment mimics its transforms to publish a gradio app that has no fastai dependencies.
-
-## Train
-
-```bash
-conda env create -f environment.yml
-```
-
-```bash
-conda activate fastai
-cd training
-python -m birds.train
-```

environment.yml

@@ -1,25 +0,0 @@
-name: fastai
-channels:
-  - pytorch
-  - nvidia
-  - conda-forge
-  - fastchan
-  - defaults
-dependencies:
-  - python=3.10
-  - pip
-  - cudatoolkit=11.7
-  - pytorch=1.13.1
-  - torchvision=0.14.1
-  - pytorch-cuda=11.7
-  - fastcore=1.5.28
-  - fastai=2.7.11
-  - pip:
-      - ipykernel
-      - ipywidgets
-      - gradio==3.20.1
-      - timm==0.6.12
-      - kaggle==1.5.12
-      - flake8
-      - black
-      - pytest

models/vit_exported.pkl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c96e179026917191d7a7e8b62c64d1a212ee77779965523a09a4b26fd6ccb934
-size 23232463

tests/test_validation_transforms.py

@@ -1,109 +0,0 @@
-import pytest
-
-from pathlib import Path
-from typing import List
-import numpy as np
-from PIL import Image
-import torch
-import torchvision.transforms as tvtfms
-from fastai.vision.data import PILImage
-import fastai.vision.augment as fastai_aug
-
-from deployment.transforms import resized_crop_pad, gpu_crop
-
-DATA_PATH = "data/200-bird-species-with-11788-images"
-
-
-def get_birds_images(path: Path) -> List[str]:
-    with open(path / "images.txt", "r") as file:
-        lines = [
-            path.resolve() / "images" / line.strip().split()[1]
-            for line in file.readlines()
-        ]
-    return lines
-
-
-class TestTransforms:
-    im_idx = 510
-
-    @pytest.fixture
-    def img_paths(self) -> List[str]:
-        path = Path(DATA_PATH) / "CUB_200_2011"
-        return get_birds_images(path.resolve())
-
-    @pytest.fixture
-    def im_fastai(self, img_paths: List[str]) -> PILImage:
-        fname = img_paths[self.im_idx]
-        return PILImage.create(fname)
-
-    @pytest.fixture
-    def im_pil(self, img_paths: List[str]) -> Image:
-        fname = img_paths[self.im_idx]
-        return Image.open(fname)
-
-    def testImageFastaiEqualsPillow(self, im_fastai: PILImage, im_pil: Image):
-        assert (np.array(im_fastai) == np.array(im_pil)).all()
-
-    # RandomResizedCrop is not exactly equal to CropPad in validation
-    # # def testRandomResizedCropEqualsCropPad(self, im_fastai: PILImage):
-    # #     crop_fastai = fastai_aug.CropPad((460, 460))
-    # #     crop_rrc = fastai_aug.RandomResizedCrop((460, 460))
-
-    # #     cropped_rrc = crop_rrc(im_fastai, split_idx=1)
-    # #     cropped_fastai = crop_fastai(im_fastai, split_idx=1)
-
-    # #     assert (np.array(cropped_rrc) == np.array(cropped_fastai)).all()
-
-    def testRandomResizedCropEqualsCustomResizedCropPad(
-        self, im_fastai: PILImage, im_pil: Image
-    ):
-        crop_rrc = fastai_aug.RandomResizedCrop((460, 460))
-
-        assert (
-            np.array(crop_rrc(im_fastai, split_idx=1))
-            == np.array(resized_crop_pad(im_pil, (460, 460)))
-        ).all()
-
-    def testFlipEqualsCustomGPUCrop(self, im_fastai: PILImage, im_pil: Image):
-        # apply flip augmentation on validation
-        tt_fastai = fastai_aug.ToTensor()
-        i2f_fastai = fastai_aug.IntToFloatTensor()
-        flip = fastai_aug.Flip(size=(224, 224))
-        result_im_fastai = flip(
-            i2f_fastai(tt_fastai(im_fastai).unsqueeze(0)), split_idx=1
-        )
-
-        # apply custom gpu crop
-        tt_torch = tvtfms.ToTensor()
-        result_im_tv = gpu_crop(tt_torch(im_pil).unsqueeze(0), size=(224, 224))
-
-        assert torch.allclose(result_im_fastai, result_im_tv)
-        assert (result_im_fastai == result_im_tv).all()
-
-    def testFastaiTransformsEqualsCustom(self, im_fastai: PILImage, im_pil: Image):
-        # fastai transforms
-        crop_rrc = fastai_aug.RandomResizedCrop((460, 460))
-        tt_fastai = fastai_aug.ToTensor()
-        i2f_fastai = fastai_aug.IntToFloatTensor()
-        flip = fastai_aug.Flip(size=(224, 224))
-        brightness = fastai_aug.Brightness()
-        norm_fastai = fastai_aug.Normalize.from_stats(
-            *fastai_aug.imagenet_stats, cuda=False
-        )
-
-        # custom transforms
-        tt_torch = tvtfms.ToTensor()
-        norm_torch = tvtfms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
-
-        # apply all fastai augmentations on validation (transformations)
-        batch_im_fastai = tt_fastai(crop_rrc(im_fastai, split_idx=1)).unsqueeze(0)
-        result_im_fastai = norm_fastai(
-            brightness(flip(i2f_fastai(batch_im_fastai), split_idx=1), split_idx=1)
-        )
-
-        # apply all custom transformations
-        batch_im_tv = tt_torch(resized_crop_pad(im_pil, (460, 460))).unsqueeze(0)
-        result_im_tv = norm_torch(gpu_crop(batch_im_tv, size=(224, 224)))
-
-        assert torch.allclose(result_im_fastai, result_im_tv)
-        assert (result_im_fastai == result_im_tv).all()

training/birds/config.py

@@ -1,6 +0,0 @@
-DATA_STORAGE_PATH = "../data"
-MODELS_STORAGE_PATH = "../models"
-DATASET = "200-bird-species-with-11788-images"
-OWNER = "veeralakrishna"
-
-DATA_PATH = DATA_STORAGE_PATH + "/" + DATASET

training/birds/train.py

@@ -1,81 +0,0 @@
-import os
-from pathlib import Path
-
-from fastai.vision.data import (
-    IndexSplitter,
-    DataBlock,
-    ImageBlock,
-    CategoryBlock,
-    RegexLabeller,
-)
-from fastai.vision.augment import (
-    RandomResizedCrop,
-    aug_transforms,
-    Normalize,
-    imagenet_stats,
-)
-
-from fastai.callback import schedule  # noqa: F401
-from fastai.vision.learner import vision_learner, accuracy
-
-from birds import config
-from birds.utils.kaggle import download_dataset
-
-
-def get_birds_images(path):
-    with open(path / "images.txt", "r") as file:
-        lines = [
-            path.resolve() / "images" / line.strip().split()[1]
-            for line in file.readlines()
-        ]
-    return lines
-
-
-def BirdsSplitter(path):
-    with open(path / "train_test_split.txt", "r") as file:
-        valid_idx = [
-            int(line.strip().split()[0]) - 1
-            for line in file.readlines()
-            if line.strip().split()[1] == "1"
-        ]
-    return IndexSplitter(valid_idx)
-
-
-if __name__ == "__main__":
-    bs = 64
-
-    if download_dataset(config.OWNER, config.DATASET, config.DATA_PATH):
-        import tarfile
-
-        with tarfile.open(Path(config.DATA_PATH) / "CUB_200_2011.tgz", "r:gz") as tar:
-            tar.extractall(path=config.DATA_PATH)
-
-        os.remove(Path(config.DATA_PATH) / "CUB_200_2011.tgz")
-        os.remove(Path(config.DATA_PATH) / "segmentations.tgz")
-
-    path = Path(config.DATA_PATH) / "CUB_200_2011"
-
-    item_tfms = RandomResizedCrop(460, min_scale=0.75, ratio=(1.0, 1.0))
-    batch_tfms = [
-        *aug_transforms(size=224, max_warp=0),
-        Normalize.from_stats(*imagenet_stats),
-    ]
-
-    birds = DataBlock(
-        blocks=(ImageBlock, CategoryBlock),
-        get_items=get_birds_images,
-        splitter=BirdsSplitter(path),
-        get_y=RegexLabeller(pat=r"/([^/]+)_\d+_\d+\.jpg"),
-        item_tfms=item_tfms,
-        batch_tfms=batch_tfms,
-    )
-
-    dls = birds.dataloaders(path)
-
-    learner = vision_learner(dls, "vit_tiny_patch16_224", metrics=[accuracy])
-
-    learner.fine_tune(7, base_lr=0.001, freeze_epochs=12)
-
-    learner.export(Path(config.MODELS_STORAGE_PATH).resolve() / "vit_exported.pkl")
-    learner.model_dir = Path(config.MODELS_STORAGE_PATH).resolve()
-    learner.save("vit_saved", with_opt=False)

training/birds/utils/kaggle.py

@@ -1,57 +0,0 @@
-import os
-from pathlib import Path
-from zipfile import ZipFile
-from kaggle import api
-
-
-def running_on_kaggle() -> bool:
-    """
-    Checks if script is running on kaggle
-    :return: true if the script is running on kaggle, false otherwise
-    """
-    if os.environ.get("KAGGLE_KERNEL_RUN_TYPE", ""):
-        return True
-    else:
-        return False
-
-
-def download_competition_data(competition: str, input_path: str | Path) -> None:
-    """
-    Downloads data from kaggle competition only if input folder is empty
-    :param comptetition: string with the competition name id of kaggle
-    :param input_path: path of the input folder
-    """
-    data_path = Path(input_path)
-    if not data_path.exists():
-        data_path.mkdir(parents=True)
-    if not any(data_path.iterdir()):
-        api.competition_download_cli(competition, path=data_path)
-        with ZipFile(data_path / (competition + ".zip"), "r") as zipObj:
-            # Extract all the contents of zip file in current directory
-            zipObj.extractall(path=data_path)
-        os.remove(data_path / (competition + ".zip"))
-
-        print(os.listdir(data_path))
-
-
-def download_dataset(owner: str, dataset: str, input_path: str | Path) -> bool:
-    """
-    Downloads data from kaggle competition only if input folder is empty
-    :param comptetition: string with the competition name id of kaggle
-    :param input_path: path of the input folder
-    """
-    downloaded = False
-    data_path = Path(input_path)
-    if not data_path.exists():
-        data_path.mkdir(parents=True)
-    if not any(data_path.iterdir()):
-        downloaded = True
-        api.dataset_download_files(f"{owner}/{dataset}", path=data_path)
-        with ZipFile(data_path / (dataset + ".zip"), "r") as zipObj:
-            # Extract all the contents of zip file in current directory
-            zipObj.extractall(path=data_path)
-        os.remove(data_path / (dataset + ".zip"))
-
-        print(os.listdir(data_path))
-
-    return downloaded