lets go!

Author: elsampsa

README.md

@@ -1,24 +1,72 @@
 # CompressAI-Vision
 
-Extensions on top of CompressAI to support video coding for machine applications
+*for complete documentation, please go directly to (TODO: link)*
+
+## Synopsis
+
+CompressAI-Vision helps you to develop, test and evaluate [CompressAI](https://interdigitalinc.github.io/CompressAI) models with standardized tests.
+
+Image and video coding models can be tested using various metrics, say SSIM and PNSR, but also againts image detection and segmentation tasks.  
+
+Developing optimized encoders for pipelines including deep-learning-based detectors is called "video coding for machines" (VCM) and its goal is to create efficient encoders for machine learning tasks:
+```
+Video stream --> Encoding + Decoding --> Detector
+```
+
+A typical metric for evaluating the encoder's efficiency for serving a detection/segmentation task, is the mean average precision (mAP) as function of a set of encoding/quality parameters:
+
+TODO: add a figure here
+
+For testing the VCM pipeline, various mAP measures (as in COCO or OpenImageV6 evaluation protocols) and datasets can be used, while the deep-learning CompressAI-based models are typically compared againts a well-known "anchor" pipeline, featuring a classical/standard image/video codec, say, H266.
+
+The **MPEG working group for VCM** has created a set of standards for testing VCM, including standardized data/image sets (typically OpenImageV6 subsets), evaluation protocols (OpenImageV6) and anchor pipelines (VTM/H266 based)
+
+## Features
+
+CompressAI-Vision makes the handling and evaluation of the mentioned datasets with any encoding/decoding and detection pipeline a breeze:
+
+- It uses [fiftyone](https://voxel51.com/docs/fiftyone/) for dataset downloading, handling, visualization and evaluation protocols.  Several evaluation protocols are supported via fiftyone
+
+- Supports CompressAI, VTM/H266 (TODO: link) and custom modules for the encoding/decoding part
+
+- Uses [Detectron2](https://detectron2.readthedocs.io/en/latest/index.html) for detector and image segmentation models
+
+- Supports official MPEG committee input files (TODO: link to mpeg vcm document?)
+
+- Single-shot CLI commands for fast input file import, image download and evaluation
+
+- Docker images, including all required components (CompressAI, VTM, Detectron2, CUDA support and whatnot) are provided to get you started asap
+
+Tutorials, API documentation and notebook examples are provided, so please, go to the documentation to get started (TODO: link here)
 
 ## Installation
 
-CompressAI-Vision supports python 3.6+ and PyTorch 1.7+.
+Your software stack will look like this (all with CUDA support):
 
+- [PyTorch](https://pytorch.org/)
+- [CompressAI](https://interdigitalinc.github.io/CompressAI)
+- [Detectron2](https://detectron2.readthedocs.io/en/latest/index.html)
+- [fiftyone](https://voxel51.com/docs/fiftyone/)
+- VTM (TODO: link)
+- _This_ library (CompressAI-Vision)
+- Jupyter notebooks
 
-**From source**:
+Instructions for creating a correct ``virtualenv`` are provided in the documentation (TODO: link).
 
-A C++17 compiler, a recent version of pip (19.0+), and common python packages
-are also required (see `setup.py` for the full list).
+Docker images with the correct software stack are also provided (TODO: link).
 
-To get started locally and install the development version of CompressAI, run
-the following commands in a [virtual environment](https://docs.python.org/3.6/library/venv.html):
+## Copyright
 
-```bash
-git clone https://github.com/InterDigitalInc/compressai-vision compressai-vision
-cd compressai-vcm
-pip install -U pip && pip install -e .
-```
+InterDigital 2022
+
+## License
+
+TODO
+
+## Authors
+
+Sampsa Riikonen
+
+Jacky Lam
 
-to be continued
+Fabien

bash/cleanpy.bash

@@ -0,0 +1,12 @@
+#!/bin/bash
+## most of these should be done through .gitignore
+## clean python bytecode
+find . -name "__pycache__" -exec rm -rf {} \;
+find . -name "*.pyc" -exec rm -rf {} \;
+## python autodoc clean:
+find . -name "*.pickle" -exec rm -rf {} \;
+## clean python build
+rm -rf dist
+rm -rf build
+# rm -rf *.egg-info
+rm -f *.deb

compressai_vision/__init__.py

@@ -0,0 +1 @@
+ 

compressai_vision/cli/__init__.py

@@ -0,0 +1,7 @@
+from compressai_vision.cli.download import main as download
+from compressai_vision.cli.list import main as list
+from compressai_vision.cli.deregister import main as deregister
+from compressai_vision.cli.register import main as register
+from compressai_vision.cli.nokia_convert import main as nokia_convert
+from compressai_vision.cli.detectron2_eval import main as detectron2_eval
+from compressai_vision.cli.load_eval import main as load_eval

compressai_vision/cli/auto.py

@@ -0,0 +1,107 @@
+"""cli.py : Command-line interface tools for compressai-vision
+
+* Copyright: 2022 InterDigital
+* Authors  : Sampsa Riikonen
+* Date     : 2022
+* Version  : 0.1
+
+This file is part of compressai-vision libraru
+"""
+import os, sys
+
+# define legit filenames.. there are inconsistencies here
+fname_list=["detection_validation_input_5k.lst",
+    "detection_validation_5k_bbox.csv", # inconsistent name
+    "detection_validation_labels_5k.csv",
+    "segmentation_validation_input_5k.lst",
+    "segmentation_validation_bbox_5k.csv",
+    "segmentation_validation_labels_5k.csv",
+    "segmentation_validation_masks_5k.csv"]
+
+help_st="""
+compressai-nokia-auto-import
+
+Automatic downloading of images and importing nokia-provided files into fiftyone
+
+Before running this command (without any arguments), put the following files into the same directory 
+(please note 'detection_validation_5k_bbox.csv' name inconsistency):
+
+"""
+for fname in fname_list:
+    help_st += fname + "\n"
+help_st += "\n"
+
+class Namespace:
+    pass
+
+def get_(key):
+    """So that we use only legit names
+    """
+    return fname_list[fname_list.index(key)]
+
+def main():
+    from compressai_vision.cli import download, nokia_convert, register
+
+    if len(sys.argv) > 1:
+        print(help_st)
+        return
+
+    for fname in fname_list:
+        if not os.path.exists(fname):
+            print("\nFATAL: missing file", fname)
+            print(help_st)
+            sys.exit(2)
+
+    print("\n**DOWNLOADING**\n")
+    p=Namespace()
+    p.y=False
+    p.name="open-images-v6"
+    p.lists=get_("detection_validation_input_5k.lst")+","+get_("segmentation_validation_input_5k.lst")
+    p.split="validation"
+    download(p)
+
+    print("\n**CONVERTING DETECTION DATA**\n")
+    p=Namespace()
+    p.y=False
+    p.lists=get_("detection_validation_input_5k.lst")
+    p.dir="~/fiftyone/open-images-v6/validation"
+    p.target_dir="~/fiftyone/nokia-detection"
+    p.label=get_("detection_validation_labels_5k.csv")
+    p.bbox=get_("detection_validation_5k_bbox.csv")
+    p.mask=None
+    nokia_convert(p)
+
+    print("\n**CONVERTING SEGMENTATION DATA**\n")
+    p=Namespace()
+    p.y=False
+    p.lists=get_("segmentation_validation_input_5k.lst")
+    p.dir="~/fiftyone/open-images-v6/validation"
+    p.target_dir="~/fiftyone/nokia-segmentation"
+    p.label=get_("segmentation_validation_labels_5k.csv")
+    p.bbox=get_("segmentation_validation_bbox_5k.csv")
+    p.mask=get_("segmentation_validation_masks_5k.csv")
+    nokia_convert(p)
+
+    print("\n**REGISTERING DETECTION DATA**\n")
+    p=Namespace()
+    p.y=False
+    p.name="nokia-detection"
+    p.lists=get_("detection_validation_input_5k.lst")
+    p.dir="~/fiftyone/nokia-detection"
+    p.type="OpenImagesV6Dataset"
+    register(p)
+
+    print("\n**REGISTERING SEGMENTATION DATA**\n")
+    p=Namespace()
+    p.y=False
+    p.name="nokia-segmentation"
+    p.lists=get_("segmentation_validation_input_5k.lst")
+    p.dir="~/fiftyone/nokia-segmentation"
+    p.type="OpenImagesV6Dataset"
+    register(p)
+    print("\nPlease continue with the compressai-vision command line tool\n")
+    print("\nGOODBYE\n")
+    
+
+if (__name__ == "__main__"):
+    main()

compressai_vision/cli/deregister.py

@@ -0,0 +1,17 @@
+"""cli deregister functionality
+""" 
+# fiftyone
+import fiftyone as fo
+import fiftyone.zoo as foz
+# compressai_vision
+from compressai_vision.conversion import imageIdFileList
+
+def main(p):
+    dataset = fo.load_dataset(p.name)
+    print("removing dataset %s from fiftyone" %(p.name))
+    if not p.y:
+        input('press enter to continue.. ')
+    try:
+        fo.delete_dataset(p.name)
+    except ValueError as e:
+        print("could not deregister because of", e)