Unified Cores and Cleanup

[kozlov721]

Unified all Core subclasses into LuxonisModel

• luxonis_train.core.Core renamed to LuxonisModel
• luxonis_train.models.LuxonisModel renamed to LuxonisLightningModule
• Instead of Trainer, Inferer, Exporter etc., there is only LuxonisModel
• Solves various issues with incorrectly created runs / creating empty runs / saving files related to one run to multiple different places etc.

Before:

from luxonis_train.core import Trainer, Exporter, Archiver

trainer = Trainer(config)
trainer.train()
exporter = Exporter(config)
exporter.export()
archiver = Archiver(config)
archiver.archive()

This would create the following directories:

.
├── output
│   ├── 0-lavender-junglefowl
│   │   ├── best_val_metric
│   │   ├── config.yaml
│   │   ├── luxonis_train.log
│   │   └── min_val_loss
│   ├── 1-peach-vole
│   ├── 2-white-gorilla
│   │   ├── config.yaml
│   │   └── luxonis_train.log
│   └── 3-cyan-sawfish
│       ├── config.yaml
│       └── luxonis_train.log
├── output_archive
│   └── nn_archive_onnx.tar.xz
└── output_export
    └── model.onnx

Now:

from luxonis_train.core import LuxonisModel

model = LuxonisModel(config)
model.train()
model.export()
model.archive()

./output/0-lavender-junglefowl/
├── archive
│   └── model.onnx.tar.xz
├── export
│   ├── model.onnx
│   └── model.yaml
├── config.yaml
├── luxonis_train.log
├── best_val_metric
└── min_val_loss

Miscellaneous

• added type hints to luxonis_ml.nodes.blocks
• reusing functions from luxonis_ml instead of defining them again
• removed unused config values
• renamed BaseLoader.input_shape to BaseLoader.input_shapes
• added upload_to_run option to exporter and tracker configuration sections
  • True by default
  • uploads the generated files as artifacts to the current run

[github-actions]

Test Results

  3 files    3 suites   32m 45s ⏱️
 51 tests  49 ✅ 0 💤 2 ❌
132 runs  128 ✅ 1 💤 3 ❌

For more details on these failures, see this check.

Results for commit 491bf0d.

♻️ This comment has been updated with latest results.