Update installation method

README.md

@@ -36,19 +36,18 @@ EoMT with DINOv2 is also available on [Hugging Face Transformers](https://huggin
 
 ## Installation
 
-If you don't have Conda installed, install Miniconda and restart your shell:
+If you don't have Pixi installed, install Pixi following the instructions in https://pixi.prefix.dev/latest/installation/ .
+
+Then install the dependencies:
 
 ```bash
-wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
-bash Miniconda3-latest-Linux-x86_64.sh
+pixi install
 ```
 
-Then create the environment, activate it, and install the dependencies:
+To activate the environment run:
 
 ```bash
-conda create -n eomt python==3.13.2
-conda activate eomt
-python3 -m pip install -r requirements.txt
+pixi s
 ```
 
 [Weights & Biases](https://wandb.ai/) (wandb) is used for experiment logging and visualization. To enable wandb, log in to your account:
@@ -98,7 +97,7 @@ wget --load-cookies cookies.txt --content-disposition https://www.cityscapes-dat
 To train EoMT from scratch, run:
 
 ```bash
-python3 main.py fit \
+pixi run main fit \
   -c configs/dinov2/coco/panoptic/eomt_large_640.yaml \
   --trainer.devices 4 \
   --data.batch_size 4 \
@@ -129,7 +128,7 @@ To fine-tune a pre-trained EoMT model, add:
 To evaluate a pre-trained EoMT model, run:
 
 ```bash
-python3 main.py validate \
+pixi run main validate \
   -c configs/dinov2/coco/panoptic/eomt_large_640.yaml \
   --model.network.masked_attn_enabled False \
   --trainer.devices 4 \

models/eomt.py → eomt/models/eomt.py

@@ -12,7 +12,7 @@
 import torch.nn.functional as F
 import math
 
-from models.scale_block import ScaleBlock
+from eomt.models.scale_block import ScaleBlock
 
 
 class EoMT(nn.Module):

training/lightning_module.py → eomt/training/lightning_module.py

@@ -34,7 +34,7 @@
 from torchvision.transforms.v2.functional import pad
 import logging
 
-from training.two_stage_warmup_poly_schedule import TwoStageWarmupPolySchedule
+from eomt.training.two_stage_warmup_poly_schedule import TwoStageWarmupPolySchedule
 
 bold_green = "\033[1;32m"
 reset = "\033[0m"

training/mask_classification_instance.py → eomt/training/mask_classification_instance.py

@@ -12,8 +12,8 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
-from training.mask_classification_loss import MaskClassificationLoss
-from training.lightning_module import LightningModule
+from eomt.training.mask_classification_loss import MaskClassificationLoss
+from eomt.training.lightning_module import LightningModule
 
 
 class MaskClassificationInstance(LightningModule):

training/mask_classification_panoptic.py → eomt/training/mask_classification_panoptic.py

@@ -8,8 +8,8 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
-from training.mask_classification_loss import MaskClassificationLoss
-from training.lightning_module import LightningModule
+from eomt.training.mask_classification_loss import MaskClassificationLoss
+from eomt.training.lightning_module import LightningModule
 
 
 class MaskClassificationPanoptic(LightningModule):

training/mask_classification_semantic.py → eomt/training/mask_classification_semantic.py

@@ -8,8 +8,8 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
-from training.mask_classification_loss import MaskClassificationLoss
-from training.lightning_module import LightningModule
+from eomt.training.mask_classification_loss import MaskClassificationLoss
+from eomt.training.lightning_module import LightningModule
 
 
 class MaskClassificationSemantic(LightningModule):

pixi.toml

@@ -0,0 +1,51 @@
+[workspace]
+channels = ["https://prefix.dev/conda-forge"]
+preview = ["pixi-build"]
+platforms = ["linux-64"]
+requires-pixi = ">=0.62"
+
+[system-requirements]
+linux = "5.4.0"
+
+[feature.cuda.system-requirements]
+cuda = "12"
+
+[package]
+name = "eomt"
+version = "0.0.1"
+description = "Your ViT is Secretly an Image Segmentation Model"
+authors = ["Smart Robotics <info@smart-robotics.nl>"]
+
+[package.build]
+backend = { name = "pixi-build-python", version = "0.4.*" }
+
+[package.host-dependencies]
+python_abi = { version = "3.12.*", build = "*_cp312" }
+hatchling = "*"
+
+[package.run-dependencies]
+timm = ">=1.0.15,<2"
+lightning = ">=2.5,<3"
+transformers = ">=4.56"
+pytorch = ">=2.7,<3"
+torchvision = ">=0.22,<1"
+torchmetrics = ">=1.7,<2"
+scipy = ">=1.15"
+matplotlib-base = ">=3.10"
+fvcore = ">=0.1.5"
+pycocotools = ">=2.0"
+gitignore-parser = ">=0.1"
+jsonargparse = ">=4.38"
+wandb = ">=0.19,<1"
+
+[dependencies]
+eomt = { path = "." }
+
+# Dev-environment-only extras: needed to run the inference notebook and
+# typed-config tooling. Kept out of [package.run-dependencies] so downstream
+# consumers of the conda package don't pull them in transitively.
+ipykernel = ">=6.29"
+typeshed-client = ">=2.8"
+
+[tasks]
+main = "python3 main.py"

pyproject.toml

@@ -0,0 +1,27 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "eomt"
+version = "0.0.1"
+description = "Your ViT is Secretly an Image Segmentation Model"
+requires-python = ">=3.12"
+dependencies = [
+    "timm>=1.0.15,<2",
+    "lightning>=2.5,<3",
+    "transformers>=4.56",
+    "torch>=2.7,<3",
+    "torchvision>=0.22,<1",
+    "torchmetrics>=1.7,<2",
+    "scipy>=1.15",
+    "matplotlib>=3.10",
+    "fvcore>=0.1.5",
+    "pycocotools>=2.0",
+    "gitignore-parser>=0.1",
+    "jsonargparse[signatures]>=4.38",
+    "wandb>=0.19",
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["eomt"]