new image setups + docs

Dockerfile_kraken

@@ -0,0 +1,40 @@
+# For NVIDIA acceleration make sure to
+# enable the NVIDIA container toolkit
+# ubuntu/jammy is the default image,
+# nvidia/cuda is the old nvidia image
+# pytorch the newer pytorch image which
+# might conflict with a tensorflow Install
+# if acceleration is desired
+
+#FROM ubuntu/jammy
+#FROM nvidia/cuda:12.1.0-cudnn8-runtime-ubuntu22.04
+FROM nvidia/cuda:12.6.3-cudnn-runtime-ubuntu22.04
+#FROM pytorch/pytorch:2.3.1-cuda12.1-cudnn8-devel
+
+# copy package content
+COPY environment_kraken.yml .
+
+# Install base utilities
+RUN apt-get update
+RUN apt-get install -y build-essential wget software-properties-common git
+
+# install libraries
+RUN apt-get install -y libgl1 libavcodec-dev libavformat-dev libswscale-dev \
+ libgstreamer-plugins-base1.0-dev libgstreamer1.0-dev \
+ libgtk2.0-dev libgtk-3-dev libpng-dev libjpeg-dev \
+ libopenexr-dev libtiff-dev libwebp-dev
+
+# install miniconda
+ENV CONDA_DIR /opt/conda
+RUN wget --quiet https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
+RUN /bin/bash ~/miniconda.sh -b -p /opt/conda
+
+# recreate and activate the environment
+# suppress TF log level output
+RUN /opt/conda/bin/conda env create -f environment_kraken.yml
+RUN echo "source activate weahtr" > ~/.bashrc
+ENV PATH $CONDA_DIR/bin:$PATH
+
+# Set the working directory on start
+# assumes that people follow the directions!
+WORKDIR /data

README.md

@@ -70,16 +70,22 @@ involved. Once build locally no further downloads will be required.
 ```bash
 docker build -f Dockerfile -t weahtr .
 ```
+
+The default install above provides support for [PyLaia](https://github.com/jpuigcerver/PyLaia)
+and [Tesseract](https://tesseract-ocr.github.io/). If you want support for the
+[Kraken](https://kraken.re/main/index.html) environment use the following code:
+
+```bash
+docker build -f Dockerfile_kraken -t weahtr .
+```
+
 > [!NOTE]
-> Repeatedly building a docker image can result in a large cache being created
-> easily 10x the data of the image (which in itself is multiple GB in size).
->
-> If you find you are running out of storage space out of the blue, check the
-> docker build cache, and prune it.
-> 
+> Both the PyLaia and Kraken environments support various open source models
+> You can list all available Kraken models by using the command line:
 > ```bash
-> docker buildx prune -f
+> kraken list
 > ```
+> PyLaia models can be found on [HugginFace](https://huggingface.co/Teklia).
 
 Make sure to have interfacing libraries running, when relying on different 
 docker base images.
@@ -101,6 +107,17 @@ For independent installs using conda
 conda env create -f environment.yml
 ```
 
+> [!NOTE]
+> Repeatedly building a docker image can result in a large cache being created
+> easily 10x the data of the image (which in itself is multiple GB in size).
+>
+> If you find you are running out of storage space out of the blue, check the
+> docker build cache, and prune it.
+> 
+> ```bash
+> docker buildx prune -f
+> ```
+
 ### Loading the package locally
 
 For now, no online `pip` based install is supported. You can install the package 

environment.yml

@@ -1,11 +1,12 @@
 name: weahtr
 channels:
-  - conda-forge
-#  - pytorch-nightly
+  - conda-forge # to be OSS stay away from default
+
+# functional PyLaia setup
 
 dependencies:
 # General python settings
-  - python==3.8
+  - python
   - conda
   - wheel
   - pip
@@ -28,10 +29,11 @@ dependencies:
   - datasets
   - transformers
   - jiwer
+  - shapely
+  - coremltools
 # OCR components
   - pytesseract
   - tesseract
-  - mittagessen::kraken
   - pip:
      - evaluate
      - torch

environment_kraken.yml

@@ -0,0 +1,54 @@
+name: weahtr
+channels:
+  - conda-forge # to be OSS stay away from default
+
+# Dependencies taken from the kraken repo:
+# https://github.com/mittagessen/kraken/blob/main/environment_cuda.yml
+
+dependencies:
+# General python settings
+  - python>=3.9
+  - wheel
+  - pip
+  - python-bidi~=0.6.0
+  - lxml
+  - regex
+  - requests
+  - pyyaml
+  - click>=8.1
+  - numpy~=1.23
+  - pillow>=9.2.0
+  - scipy~=1.13.0
+  - jinja2~=3.0
+  - conda-forge::torchvision>=0.5.0
+  - conda-forge::pytorch~=2.4.0
+  - cudatoolkit>=9.2
+  - jsonschema
+  - scikit-learn~=1.2.1
+  - scikit-image~=0.24.0
+  - shapely>=2.0.6
+  - pyvips
+  - imagemagick>=7.1.0
+  - pyarrow
+  - importlib-resources>=1.3.0
+  - conda-forge::lightning~=2.4.0
+  - conda-forge::torchmetrics>=1.1.0
+  - conda-forge::threadpoolctl~=3.5.0
+  - pip
+  - albumentations
+  - rich
+  - setuptools>=36.6.0,<70.0.0
+  - transformers
+  - jiwer
+  - datasets
+  - tiktoken
+  - opencv
+  - pandas
+  - matplotlib
+  - pytesseract
+  - tesseract
+  - pip:
+    - coremltools~=8.1
+    - htrmopo
+    - platformdirs
+    - kraken