Deepforest-pytorch uses a config.yml to control hyperparameters related to model training and evaluation. This allows all the relevant parameters to live in one location and be easily changed when exploring new models.
Deepforest-pytorch comes with a sample config file, deepforest_config.yml. Edit this file to change settings while developing models.
# Config file for DeepForest-pytorch module
#cpu workers for data loaders
#Dataloaders
workers: 1
gpus:
distributed_backend:
batch_size: 1
#Non-max supression of overlapping predictions
nms_thresh: 0.05
score_thresh: 0.1
train:
csv_file:
root_dir:
#Optomizer initial learning rate
lr: 0.001
#Print loss every n epochs
epochs: 1
#Useful debugging flag in pytorch lightning, set to True to get a single batch of training to test settings.
fast_dev_run: False
validation:
#callback args
csv_file:
root_dir:
#Intersection over union evaluation
iou_threshold: 0.4
Number of workers to perform asynchronous data generation during model training. Corresponds to num_workers in pytorch base class https://pytorch.org/docs/stable/data.html. To turn off asynchronous data generation set workers = 0.
The number of gpus to use during model training. To run on cpu leave blank. Deepforest-pytorch has been tested on up to 8 gpu and follows a pytorch lightning module, which means it can inherent any of the scaling functionality from this library, including TPU support. https://pytorch-lightning.readthedocs.io/en/stable/advanced/multi_gpu.html?highlight=multi%20gpu
Data parallelization strategy from https://pytorch-lightning.readthedocs.io/en/stable/advanced/multi_gpu.html?highlight=multi%20gpu. Default is 'ddp' distributed data parallel, which splits data into pieces and each GPU gets a mutaully exclusive piece. Weights are updated after each forward training pass. This is most appropriate for SLURM clusters where the GPUs may be on different nodes.
Number of images per batch during training. GPU memory limits this usually between 5-10
Non-max suppression threshold. The higher scoring predicted box is kept when predictions overlap by greater than nms_thresh. For details see https://pytorch.org/vision/stable/ops.html#torchvision.ops.nms
Score threshold of predictions to keep. Predictions with less than this threshold are removed from output.
Path to csv_file for training annotations. Annotations are .csv files with headers image_path, xmin, ymin, xmax, ymax, label. image_path are relative to the root_dir. For example this file should have entries like myimage.tif not /path/to/myimage.tif
Directory to search for images in the csv_file image_path column
Learning rate for the training optimization. By default the optimizer is stochastic gradient descent with momentum. A learning rate scheduler is used based on validation loss
optim.SGD(self.model.parameters(), lr=self.config["train"]["lr"], momentum=0.9)
self.scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(self.optimizer, mode='min',
factor=0.1, patience=10,
verbose=True, threshold=0.0001,
threshold_mode='rel', cooldown=0,
min_lr=0, eps=1e-08)
This scheduler can be overwritten by replacing the model class
m = main.deepforest()
m.scheduler = <>
The number of times to run a full pass of the dataloader during model training.
A useful pytorch lightning flag that will run a debug run to test inputs. See
Optional validation dataloader to run during training.
Path to csv_file for validation annotations. Annotations are .csv files with headers image_path, xmin, ymin, xmax, ymax, label. image_path are relative to the root_dir. For example this file should have entries like myimage.tif not /path/to/myimage.tif
Directory to search for images in the csv_file image_path column