CenterNet: Objects as Points paper

.vscode/settings.json

@@ -1,3 +1,3 @@
 {
-  "python.pythonPath": "/usr/bin/python3.8"
+  "python.pythonPath": "C:\\Users\\david\\AppData\\Local\\Programs\\Python\\Python37\\python.exe"
 }

centernet/README.md

@@ -0,0 +1,130 @@
+> :memo: A README.md template for releasing a paper code implementation to a GitHub repository.
+>
+> * Template version: 1.0.2020.170
+> * Please modify sections depending on needs.
+
+# CenterNet
+
+[![Paper](http://img.shields.io/badge/Paper-arXiv.1904.07850-B3181B?logo=arXiv)](https://arxiv.org/abs/1904.07850)
+
+This repository is the unofficial implementation of the following paper.
+
+* Paper title: [Objects as Points](https://arxiv.org/abs/1904.07850)
+
+## Description
+
+CenterNet [1] builds upon CornerNet [2], an anchor-free model for object
+detection.
+
+Many other models, such as YOLO and RetinaNet, use anchor boxes. These anchor
+boxes are predefined to be close to the aspect ratios and scales of the objects
+in the training dataset. Anchor-based models do not predict the bounding boxes
+of objects directly. They instead predict the location and size/shape
+refinements to a predefined anchor box. The detection generator then computes
+the final confidences, positions, and size of the detection.
+
+CornerNet eliminates the need for anchor boxes. RetinaNet needs thousands of
+anchor boxes in order to cover the most common ground truth boxes [2]. This adds
+unnecessary complexity to the model which slow down training and create
+imbalances in positive and negative anchor boxes [2]. Instead, CornerNet creates
+heatmaps for each of the corners and pools them together in order to get the
+final detection boxes for the objects. CenterNet removes even more complexity
+by using the center instead of the corners, meaning that only one set of
+heatmaps (one heatmap for each class) is needed to predict the object. CenterNet
+proves that this can be done without a significant difference in accuracy.
+
+## History
+
+> :memo: Provide a changelog.
+
+## Authors or Maintainers
+
+> :memo: Provide maintainer information.
+
+* Full name ([@GitHub username](https://github.com/username))
+* Full name ([@GitHub username](https://github.com/username))
+
+## Table of Contents
+
+> :memo: Provide a table of contents to help readers navigate a lengthy README document.
+
+## Requirements
+
+[![TensorFlow 2.4](https://img.shields.io/badge/TensorFlow-2.4-FF6F00?logo=tensorflow)](https://github.com/tensorflow/tensorflow/releases/tag/v2.4.0)
+[![Python 3.7](https://img.shields.io/badge/Python-3.7-3776AB)](https://www.python.org/downloads/release/python-370/)
+
+```setup
+pip install -r requirements.txt
+```
+
+## Results
+
+[![TensorFlow Hub](https://img.shields.io/badge/TF%20Hub-Models-FF6F00?logo=tensorflow)](https://tfhub.dev/...)
+
+> :memo: Provide a table with results. (e.g., accuracy, latency)
+>
+> * Provide links to the pre-trained models (checkpoint, SavedModel files).
+>   * Publish TensorFlow SavedModel files on TensorFlow Hub (tfhub.dev) if possible.
+> * Add links to [TensorBoard.dev](https://tensorboard.dev/) for visualizing metrics.
+>
+> An example table for image classification results
+>
+> ### Image Classification
+>
+> | Model name | Download | Top 1 Accuracy | Top 5 Accuracy |
+> |------------|----------|----------------|----------------|
+> | Model name | [Checkpoint](https://drive.google.com/...), [SavedModel](https://tfhub.dev/...) | xx% | xx% |
+
+## Dataset
+
+> :memo: Provide information of the dataset used.
+
+## Training
+
+> :memo: Provide training information.
+>
+> * Provide details for preprocessing, hyperparameters, random seeds, and environment.
+> * Provide a command line example for training.
+
+Please run this command line for training.
+
+```shell
+python3 ...
+```
+
+## Evaluation
+
+> :memo: Provide an evaluation script with details of how to reproduce results.
+>
+> * Describe data preprocessing / postprocessing steps.
+> * Provide a command line example for evaluation.
+
+Please run this command line for evaluation.
+
+```shell
+python3 ...
+```
+
+## References
+
+[1] CenterNet: Xingyi Zhou, Dequan Wang, and Philipp Krähenbühl. Objects as Points. arXiv preprint arXiv:1904.07850, 2019. (https://arxiv.org/abs/1904.07850)
+[2] CornerNet: Hei Law and Jia Deng. CornerNet: Detecting Objects as Paired Keypoints. arXiv preprint arXiv:1808.01244, 2018. (https://arxiv.org/abs/1808.01244)
+
+## License
+
+[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+
+> :memo: Place your license text in a file named LICENSE in the root of the repository.
+>
+> * Include information about your license.
+> * Reference: [Adding a license to a repository](https://help.github.com/en/github/building-a-strong-community/adding-a-license-to-a-repository)
+
+This project is licensed under the terms of the **Apache License 2.0**.
+
+## Citation
+
+> :memo: Make your repository citable.
+>
+> * Reference: [Making Your Code Citable](https://guides.github.com/activities/citable-code/)
+
+If you want to cite this repository in your research paper, please use the following information.

centernet/common/registry_imports.py

@@ -14,5 +14,10 @@
 # ==============================================================================
 """All necessary imports for registration."""
 
+import centernet
+from centernet.configs import centernet
+from centernet.configs.centernet import CenterNetTask
+from centernet.modeling.backbones import hourglass
+from centernet.tasks.centernet import CenterNetTask
 # pylint: disable=unused-import
 from official.common import registry_imports

centernet/configs/backbones.py

@@ -0,0 +1,41 @@
+# Lint as: python3
+# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Backbones configurations."""
+# Import libraries
+import dataclasses
+from typing import List
+
+from official.modeling import hyperparams
+from official.vision.beta.configs import backbones
+
+
+@dataclasses.dataclass
+class Hourglass(hyperparams.Config):
+  """Hourglass config."""
+  input_channel_dims: int = 128
+  channel_dims_per_stage: List[int] = dataclasses.field(
+      default_factory=lambda: [256, 256, 384, 384, 384, 512])
+  blocks_per_stage: List[int] = dataclasses.field(
+      default_factory=lambda: [2, 2, 2, 2, 2, 4])
+  num_hourglasses: int = 2
+  initial_downsample: bool = True
+  norm_momentum: float = 0.1
+  norm_episilon: float = 1e-5
+
+
+@dataclasses.dataclass
+class Backbone(backbones.Backbone):
+  hourglass: Hourglass = Hourglass()