Back to basics

hydra_flash/conf/config.yaml

-defaults:
-  - experiment: hymenoptera

hydra_flash/conf/config1.yaml

-preprocess:
-  _target_: flash.core.data.utils.download_data
-  url: "https://pl-flash-data.s3.amazonaws.com/hymenoptera_data.zip"
-  path: "/tmp"
-
-datamodule:
-  _target_: flash.image.ImageClassificationData.from_folders
-  train_folder: ${preprocess.path}/hymenoptera_data/train/
-  val_folder: ${preprocess.path}/hymenoptera_data/val/
-  batch_size: 4
-  num_workers: 12
-  transform_kwargs:
-    image_size: [196, 196]
-    mean: [0.485, 0.456, 0.406]
-    std: [0.229, 0.224, 0.225]
-
-model:
-  _target_: flash.image.ImageClassifier
-  backbone: resnet18
-
-trainer:
-  _target_: flash.Trainer
-  max_epochs: 3
-  gpus: 1
-
-predict_datamodule:
-  _target_: flash.image.ImageClassificationData.from_files
-  predict_files:
-    - "${preprocess.path}/hymenoptera_data/val/bees/65038344_52a45d090d.jpg"
-    - "${preprocess.path}/hymenoptera_data/val/bees/590318879_68cf112861.jpg"
-    - "${preprocess.path}/hymenoptera_data/val/ants/540543309_ddbb193ee5.jpg"
-  batch_size: 3
-
-show: true
-save: false

hydra_flash/conf/config2.yaml

-defaults:
-  - _self_
-  - server: local
-  - preprocess: hymenoptera
-  - datamodule: hymenoptera
-  - model: resnet18
-  - datamodule@predict_datamodule: hymenoptera_test
-
-trainer:
-  _target_: flash.Trainer
-  max_epochs: 3
-  gpus: 1

hydra_flash/conf/datamodule/hymenoptera.yaml

-_target_: flash.image.ImageClassificationData.from_folders
-train_folder: ${preprocess.path}/hymenoptera_data/train/
-val_folder: ${preprocess.path}/hymenoptera_data/val/
-batch_size: 4
-num_workers: 12
-transform_kwargs:
-  image_size: [196, 196]
-  mean: [0.485, 0.456, 0.406]
-  std: [0.229, 0.224, 0.225]

hydra_flash/conf/datamodule/hymenoptera_test.yaml

-_target_: flash.image.ImageClassificationData.from_files
-predict_files:
-  - "${preprocess.path}/hymenoptera_data/val/bees/65038344_52a45d090d.jpg"
-  - "${preprocess.path}/hymenoptera_data/val/bees/590318879_68cf112861.jpg"
-  - "${preprocess.path}/hymenoptera_data/val/ants/540543309_ddbb193ee5.jpg"
-batch_size: 3

hydra_flash/conf/model/gernet.yaml

-_target_: flash.image.ImageClassifier
-backbone: gernet_s
-learning_rate: 1.e-3

hydra_flash/conf/model/resnet18.yaml

-_target_: flash.image.ImageClassifier
-backbone: resnet18
-learning_rate: 1.e-3

hydra_flash/conf/preprocess/download.yaml

-_target_: flash.core.data.utils.download_data
-path: ${path}

hydra_flash/conf/preprocess/hymenoptera.yaml

-defaults:
-  - download
-
-url: "https://pl-flash-data.s3.amazonaws.com/hymenoptera_data.zip"

hydra_flash/conf/server/local.yaml

-# @package _global_
-show: true
-save: true
-path: "/tmp"

hydra_flash/conf/server/mb.yaml

-# @package _global_
-defaults:
-  - override /hydra/launcher: submitit_slurm
-
-show: false
-save: true
-preprocess:
-  path: ${oc.env:TEMPDIR}
-
-gpus: [0]
-hydra:
-  launcher:
-    name: hydra_test
-    timeout_min: 360
-    cpus_per_task: 12
-    tasks_per_node: 1
-    signal_delay_s: 120
-    max_num_timeout: 20
-    mem_gb: 32
-    nodes: 1
-    array_parallelism: 4
-    partition: gpu
-    gres: "gpu:1g.10gb:1"

hydra_flash/train.py

-import hydra
-from hydra.utils import call, instantiate as hydra_instantiate
-from functools import partial
-from omegaconf import OmegaConf
 import logging
+import torch
 
+import flash
+from flash.core.data.utils import download_data
+from flash.image import ImageClassificationData, ImageClassifier
 from hydra_flash.utils import show_predictions
 
-log = logging.getLogger(__name__)
 
-instantiate = partial(hydra_instantiate, _convert_="all")
-
-
-@hydra.main(version_base="1.1", config_path="conf", config_name="config2")
-def main(cfg):
-    # Preprocess : download and/or unzip data
-    call(cfg.preprocess)
-
-    # Create datamodule from downloaded files
-    datamodule = instantiate(cfg.datamodule)
+def main():
+    # 1. Create the DataModule
+    download_data(
+        "https://pl-flash-data.s3.amazonaws.com/hymenoptera_data.zip", "./data"
+    )
 
-    # Create model
-    model = instantiate(
-        cfg.model, labels=datamodule.labels, multi_label=datamodule.multi_label
+    datamodule = ImageClassificationData.from_folders(
+        train_folder="data/hymenoptera_data/train/",
+        val_folder="data/hymenoptera_data/val/",
+        batch_size=4,
+        transform_kwargs={
+            "image_size": (196, 196),
+            "mean": (0.485, 0.456, 0.406),
+            "std": (0.229, 0.224, 0.225),
+        },
     )
 
-    # Finetune model
-    trainer = instantiate(cfg.trainer)
-    trainer.finetune(model, datamodule=datamodule, strategy="freeze")
+    # 2. Build the task
+    model = ImageClassifier(backbone="resnet18", labels=datamodule.labels)
 
-    # Predict and show predictions
-    predict_datamodule = instantiate(cfg.predict_datamodule)
-    predictions = trainer.predict(model, datamodule=predict_datamodule, output="labels")
+    # 3. Create the trainer and finetune the model
+    trainer = flash.Trainer(max_epochs=3, gpus=torch.cuda.device_count())
+    trainer.finetune(model, datamodule=datamodule, strategy="freeze")
 
-    show_predictions(
-        predict_datamodule.predict_dataset, predictions[0], cfg.show, cfg.save
+    # 4. Predict what's on a few images! ants or bees?
+    datamodule = ImageClassificationData.from_files(
+        predict_files=[
+            "data/hymenoptera_data/val/bees/65038344_52a45d090d.jpg",
+            "data/hymenoptera_data/val/bees/590318879_68cf112861.jpg",
+            "data/hymenoptera_data/val/ants/540543309_ddbb193ee5.jpg",
+        ],
+        batch_size=3,
     )
+    predictions = trainer.predict(model, datamodule=datamodule, output="labels")
+
+    show_predictions(datamodule.predict_dataset, predictions[0], show=True, save=False)
 
 
 if __name__ == "__main__":