Skip to content

Instance segmentation better parameters for proper training #90

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dtronmans merged 2 commits into from
Mar 6, 2026
Merged

Instance segmentation better parameters for proper training #90

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7bdf0ce
Instance segmentation better parameters for proper training
dtronmans Feb 13, 2026
276a2ff
Better working parameters
dtronmans Feb 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 3 additions & 13 deletions training/train_instance_segmentation_model.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -409,7 +409,7 @@
"In this section, we can set up everything connected to the actual training. You can change preprocessing, batch size, epochs, add callbacks, augmentations, change optimizers, schedulers, and more. Please refer to the [complete documentation](https://github.com/luxonis/luxonis-train/tree/main/configs). \n",
"\n",
"#### Augmentations\n",
"In this tutorial, we'll leave most things as they are; the only change will be adding some augmentations. `Luxonis-train` uses [`Albumentations`](https://albumentations.ai/) for augmentations by adding custom ones like `Mosaic4` and `MixUp`. You can use [this demo](https://demo.albumentations.ai/) to experiment and find those that work for your specific training run. In this training run, we'll add `Affine`, `HorizontalFlip`, `ColorJitter`, `Sharpen` and `Mosaic4` augmentations, but feel free to edit this.\n",
"In this tutorial, we'll leave most things as they are; the only change will be adding some augmentations. `Luxonis-train` uses [`Albumentations`](https://albumentations.ai/) for augmentations by adding custom ones like `MixUp`. You can use [this demo](https://demo.albumentations.ai/) to experiment and find those that work for your specific training run. In this training run, we'll add `HorizontalFlip`, `ColorJitter` and `Sharpen` augmentations, but feel free to edit this.\n",
"\n",
"#### Callbacks \n",
"Callbacks are very helpful when merging more functionalities into a single training run. For example, we want to train the model, evaluate it on a test subset, export it, and create an archive. These steps can be defined through the config and done by a single call. You can check out all the available callbacks [here](https://github.com/luxonis/luxonis-train/tree/main/luxonis_train/callbacks)."
Expand Down Expand Up @@ -456,20 +456,14 @@
"\n",
"trainer:\n",
" preprocessing:\n",
" train_image_size: [640, 640]\n",
" train_image_size: [448, 448]\n",
" keep_aspect_ratio: true\n",
" normalize:\n",
" active: true\n",
" params:\n",
" mean: [0., 0., 0.]\n",
" std: [1, 1, 1]\n",
" augmentations:\n",
" - name: Affine\n",
" params:\n",
" scale: [0.7, 1.6]\n",
" rotate: 30\n",
" shear: 10\n",
" p: 0.3\n",
" - name: HorizontalFlip\n",
" params:\n",
" p: 0.3\n",
Expand All @@ -484,10 +478,6 @@
" params:\n",
" p: 0.3\n",
" - name: RandomRotate90\n",
" - name: Mosaic4\n",
" params:\n",
" out_width: 640\n",
" out_height: 640\n",
"\n",
"\n",
" batch_size: 8\n",
Expand Down Expand Up @@ -516,7 +506,7 @@
" warmup_epochs: 5\n",
" warmup_bias_lr: 0.0\n",
" warmup_momentum: 0.8\n",
" lr: 0.005\n",
" lr: 0.01\n",
" lre: 0.0001\n",
" momentum: 0.937 \n",
" weight_decay: 0.001\n",
Expand Down