Weakly Supervised Segmentation by Tensorflow.

Related tags

Deep Learningtensorflowsemantic-segmentationdeeplab-largefovweakly-supervised-learning
Overview

Simple-does-it-weakly-supervised-instance-and-semantic-segmentation

There are five weakly supervised networks in Simple Does It: Weakly Supervised Instance and Semantic Segmentation, by Khoreva et al. (CVPR 2017). Respectively, Naive, Box, Box^i, Grabcut+, M∩G+. All of them use cheap-to-generate label, bounding box, during training and don't need other informations except image during testing.

This repo contains a TensorFlow implementation of Grabcut version of semantic segmentation.

My Environment

Environment 1

  • Operating System:
    • Arch Linux 4.15.15-1
  • Memory
    • 64GB
  • CUDA:
    • CUDA V9.0.176
  • CUDNN:
    • CUDNN 7.0.5-2
  • GPU:
    • GTX 1070 8G
  • Nvidia driver:
    • 390.25
  • Python:
    • python 3.6.4
  • Python package:
    • tqdm, bs4, opencv-python, pydensecrf, cython...
  • Tensorflow:
    • tensorflow-gpu 1.5.0

Environment 2

  • Operating System:
    • Ubuntu 16.04
  • Memory
    • 64GB
  • CUDA:
    • CUDA V9.0.176
  • CUDNN:
    • CUDNN 7
  • GPU:
    • GTX 1060 6G
  • Nvidia driver:
    • 390.48
  • Python:
    • python 3.5.2
  • Python package:
    • tqdm, bs4, opencv-python, pydensecrf, cython...
  • Tensorflow:
    • tensorflow-gpu 1.6.0

Downloading the VOC12 dataset

Setup Dataset

My directory structure

./Simple_does_it/
├── Dataset
│   ├── Annotations
│   ├── CRF_masks
│   ├── CRF_pairs
│   ├── Grabcut_inst
│   ├── Grabcut_pairs
│   ├── JPEGImages
│   ├── Pred_masks
│   ├── Pred_pairs
│   ├── SegmentationClass
│   └── Segmentation_label
├── Model
│   ├── Logs
│   └── models
├── Parser_
├── Postprocess
├── Preprocess
└── Util

VOC2012 directory structure

VOCtrainval_11-May-2012
└── VOCdevkit
    └── VOC2012
        ├── Annotations
        ├── ImageSets
        │   ├── Action
        │   ├── Layout
        │   ├── Main
        │   └── Segmentation
        ├── JPEGImages
        ├── SegmentationClass
        └── SegmentationObject
  • Put annotations in 'Annotations'
mv {PATH}/VOCtrainval_11-May-2012/VOCdevkit/VOC2012/Annotations/* {PATH}/Simple_does_it/Dataset/Annotations/
  • Put images in 'JPEGImages'
mv {PATH}/VOCtrainval_11-May-2012/VOCdevkit/VOC2012/JPEGImages/* {PATH}/Simple_does_it/Dataset/JPEGImages/
  • Put Ground truth in 'SegmentationClass' for computing mIoU and IoU
mv {PATH}/VOCtrainval_11-May-2012/VOCdevkit/VOC2012/SegmentationClass/* {PATH}/Simple_does_it/Dataset/SegmentationClass/

Demo (See Usage for more details)

Download pretrain model training on VOC12 (train set size: 1464)

  • Pretrain model
    • Move files from VOC12_CKPT to 'models'
  • Run test 
    python ./Model/model.py --restore_target 1020
  • Run train (See Training for more details) 
    python ./Model/model.py --is_train 1 --set_name voc_train.txt --restore_target 1020
  • Performance
set CRF mIoU
train X 64.93%
train O 66.90%
val X 39.03%
val O 42.54%

Download pretrain model training on VOC12 + SBD (train set size: 10582)

  • Pretrain model
    • Move files from VOC12_SBD_CKPT to 'models'
  • Run test 
    python ./Model/model.py --restore_target 538
  • Run train (See Training for more details) 
    python ./Model/model.py --is_train 1 --set_name train.txt --restore_target 538
  • Performance
set CRF mIoU
train X 66.87%
train O 68.21%
val X 51.90%
val O 54.52%

Training (See Usage for more details)

Download pretrain vgg16

Extract annotations from 'Annotations' according to 'train.txt' or 'voc_train.txt' for VOC12 + SDB or VOC12

  • For VOC12 + SBD (train set size: 10582)
    • This will generate a 'train_pairs.txt' for 'grabcut.py'
    python ./Dataset/make_train.py
  • For VOC12 (train set size: 1464)
    • This will generate a 'voc_train_pairs.txt' for 'grabcut.py'
    python ./Dataset/make_train.py --train_set_name voc_train.txt --train_pair_name voc_train_pairs.txt

Generate label for training by 'grabcut.py'

  • Result of grabcut for each bounding box will be stored at 'Grabcut_inst'
  • Result of grabcut for each image will be stored at 'Segmentation_label'
  • Result of grabcut for each image combing with image and bounding box will be stored at 'Grabcut_pairs'
  • Note: If the instance hasn't existed at 'Grabcut_inst', grabcut.py will grabcut that image
  • For VOC12 + SBD (train set size: 10582) 
    python ./Preprocess/grabcut.py
  • For VOC12 (train set size: 1464) 
    python ./Preprocess/grabcut.py --train_pair_name voc_train_pairs.txt

Train network

  • The event file for tensorboard will be stored at 'Logs'
  • Train on VOC12 + SBD (train set size: 10582)
    • This will consume lot of memory.
      • The train set is so large.
      • Data dtyp will be casted from np.uint8 to np.float16 for mean substraction.
    • Eliminate mean substraction for lower memory usage.
      • Change the dtype in ./Dataset/load.py from np.float16 to np.uint8
      • Comment mean substraction in ./Model/model.py
    python ./Model/model.py --is_train 1 --set_name train.txt
  • Train on VOC12 (train set size: 1464) 
    python ./Model/model.py --is_train 1 --set_name voc_train.txt

Testing (See Usage for more details)

Test network

  • Result will be stored at 'Pred_masks'
  • Result combing with image will be stored at 'Pred_pairs'
  • Result after dense CRF will be stored at 'CRF_masks'
  • Result after dense CRF combing with image will be stored at 'CRF_pairs'
  • Test on VOC12 (val set size: 1449) 
    python ./Model/model.py --restore_target {num}

Performance (See Usage for more details)

Evaluate mIoU and IoU

  • Compute mIoU and IoU 
    python ./Dataset/mIoU.py

Usage

Parser_/parser.py

  • Parse the command line argument

Util/divied.py

  • Generating train.txt and test.txt according to 'JPEGImages'
  • Not necessary
usage: divied.py [-h] [--dataset DATASET] [--img_dir_name IMG_DIR_NAME]
                 [--train_set_ratio TRAIN_SET_RATIO]
                 [--train_set_name TRAIN_SET_NAME]
                 [--test_set_name TEST_SET_NAME]

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     path to dataset (default: Util/../Parser_/../Dataset)
  --img_dir_name IMG_DIR_NAME
                        name for image directory (default: JPEGImages)
  --train_set_ratio TRAIN_SET_RATIO
                        ratio for training set, [0,10] (default: 7)
  --train_set_name TRAIN_SET_NAME
                        name for training set (default: train.txt)
  --test_set_name TEST_SET_NAME
                        name for testing set (default: val.txt)

Dataset/make_train.py

  • Extract annotations from 'Annotations' according to 'train.txt'
  • Content: {image name}###{image name + num + class + .png}###{bbox ymin}###{bbox xmin}###{bbox ymax}###{bbox xmax}###{class}
  • Example: 2011_003038###2011_003038_3_15.png###115###1###233###136###person
usage: make_train.py [-h] [--dataset DATASET]
                     [--train_set_name TRAIN_SET_NAME]
                     [--ann_dir_name ANN_DIR_NAME]
                     [--train_pair_name TRAIN_PAIR_NAME]

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     path to dataset (default:
                        Dataset/../Parser_/../Dataset)
  --train_set_name TRAIN_SET_NAME
                        name for training set (default: train.txt)
  --ann_dir_name ANN_DIR_NAME
                        name for annotation directory (default: Annotations)
  --train_pair_name TRAIN_PAIR_NAME
                        name for training pair (default: train_pairs.txt)

Preprocess/grabcut.py

  • Grabcut a traditional computer vision method
  • Input bounding box and image then generating label for training
usage: grabcut.py [-h] [--dataset DATASET] [--img_dir_name IMG_DIR_NAME]
                  [--train_pair_name TRAIN_PAIR_NAME]
                  [--grabcut_dir_name GRABCUT_DIR_NAME]
                  [--img_grabcuts_dir IMG_GRABCUTS_DIR]
                  [--pool_size POOL_SIZE] [--grabcut_iter GRABCUT_ITER]
                  [--label_dir_name LABEL_DIR_NAME]

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     path to dataset (default:
                        ./Preprocess/../Parser_/../Dataset)
  --img_dir_name IMG_DIR_NAME
                        name for image directory (default: JPEGImages)
  --train_pair_name TRAIN_PAIR_NAME
                        name for training pair (default: train_pairs.txt)
  --grabcut_dir_name GRABCUT_DIR_NAME
                        name for grabcut directory (default: Grabcut_inst)
  --img_grabcuts_dir IMG_GRABCUTS_DIR
                        name for image with grabcuts directory (default:
                        Grabcut_pairs)
  --pool_size POOL_SIZE
                        pool for multiprocess (default: 4)
  --grabcut_iter GRABCUT_ITER
                        grabcut iteration (default: 3)
  --label_dir_name LABEL_DIR_NAME
                        name for label directory (default: Segmentation_label)

Model/model.py

  • Deeplab-Largefov
usage: model.py [-h] [--dataset DATASET] [--set_name SET_NAME]
                [--label_dir_name LABEL_DIR_NAME]
                [--img_dir_name IMG_DIR_NAME] [--classes CLASSES]
                [--batch_size BATCH_SIZE] [--epoch EPOCH]
                [--learning_rate LEARNING_RATE] [--momentum MOMENTUM]
                [--keep_prob KEEP_PROB] [--is_train IS_TRAIN]
                [--save_step SAVE_STEP] [--pred_dir_name PRED_DIR_NAME]
                [--pair_dir_name PAIR_DIR_NAME] [--crf_dir_name CRF_DIR_NAME]
                [--crf_pair_dir_name CRF_PAIR_DIR_NAME] [--width WIDTH]
                [--height HEIGHT] [--restore_target RESTORE_TARGET]

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     path to dataset (default:
                        ./Model/../Parser_/../Dataset)
  --set_name SET_NAME   name for set (default: val.txt)
  --label_dir_name LABEL_DIR_NAME
                        name for label directory (default: Segmentation_label)
  --img_dir_name IMG_DIR_NAME
                        name for image directory (default: JPEGImages)
  --classes CLASSES     number of classes for segmentation (default: 21)
  --batch_size BATCH_SIZE
                        batch size for training (default: 16)
  --epoch EPOCH         epoch for training (default: 2000)
  --learning_rate LEARNING_RATE
                        learning rate for training (default: 0.01)
  --momentum MOMENTUM   momentum for optimizer (default: 0.9)
  --keep_prob KEEP_PROB
                        probability for dropout (default: 0.5)
  --is_train IS_TRAIN   training or testing [1 = True / 0 = False] (default:
                        0)
  --save_step SAVE_STEP
                        step for saving weight (default: 2)
  --pred_dir_name PRED_DIR_NAME
                        name for prediction masks directory (default:
                        Pred_masks)
  --pair_dir_name PAIR_DIR_NAME
                        name for pairs directory (default: Pred_pairs)
  --crf_dir_name CRF_DIR_NAME
                        name for crf prediction masks directory (default:
                        CRF_masks)
  --crf_pair_dir_name CRF_PAIR_DIR_NAME
                        name for crf pairs directory (default: CRF_pairs)
  --width WIDTH         width for resize (default: 513)
  --height HEIGHT       height for resize (default: 513)
  --restore_target RESTORE_TARGET
                        target for restore (default: 0)

Dataset/mIoU.py

  • Compute mIoU and IoU
usage: mIoU.py [-h] [--dataset DATASET] [--set_name SET_NAME]
               [--GT_dir_name GT_DIR_NAME] [--Pred_dir_name PRED_DIR_NAME]
               [--classes CLASSES]

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET     path to dataset (default:
                        ./Dataset/../Parser_/../Dataset)
  --set_name SET_NAME   name for set (default: val.txt)
  --GT_dir_name GT_DIR_NAME
                        name for ground truth directory (default:
                        SegmentationClass)
  --Pred_dir_name PRED_DIR_NAME
                        name for prediction directory (default: CRF_masks)
  --classes CLASSES     number of classes (default: 21)

Dataset/load.py

  • Loading data for training / testing according to train.txt / val.txt

Dataset/save_result.py

  • Save result during testing

Dataset/voc12_class.py

  • Map the class to grayscale value

Dataset/voc12_color.py

  • Map the grayscale value to RGB

Postprocess/dense_CRF.py

  • Dense CRF a machine learning method
  • Refine the result

Reference

Owner
CHENG-YOU LU
CHENG-YOU LU
GitHub Repository
Distributed Asynchronous Hyperparameter Optimization better than HyperOpt.

UltraOpt : Distributed Asynchronous Hyperparameter Optimization better than HyperOpt. UltraOpt is a simple and efficient library to minimize expensive

98 Aug 16, 2022
The 1st place solution of track2 (Vehicle Re-Identification) in the NVIDIA AI City Challenge at CVPR 2021 Workshop.

AICITY2021_Track2_DMT The 1st place solution of track2 (Vehicle Re-Identification) in the NVIDIA AI City Challenge at CVPR 2021 Workshop. Introduction

Hao Luo 91 Dec 21, 2022
GPU-accelerated PyTorch implementation of Zero-shot User Intent Detection via Capsule Neural Networks

GPU-accelerated PyTorch implementation of Zero-shot User Intent Detection via Capsule Neural Networks This repository implements a capsule model Inten

Joel Huang 15 Dec 24, 2022
Learning Optical Flow from a Few Matches (CVPR 2021)

Learning Optical Flow from a Few Matches This repository contains the source code for our paper: Learning Optical Flow from a Few Matches CVPR 2021 Sh

Shihao Jiang (Zac) 159 Dec 16, 2022
Asynchronous Advantage Actor-Critic in PyTorch

Asynchronous Advantage Actor-Critic in PyTorch This is PyTorch implementation of A3C as described in Asynchronous Methods for Deep Reinforcement Learn

Reiji Hatsugai 38 Dec 12, 2022
Food recognition model using convolutional neural network & computer vision

Food recognition model using convolutional neural network & computer vision. The goal is to match or beat the DeepFood Research Paper

Hemanth Chandran 1 Jan 13, 2022
(CVPR2021) Kaleido-BERT: Vision-Language Pre-training on Fashion Domain

Kaleido-BERT: Vision-Language Pre-training on Fashion Domain Mingchen Zhuge*, Dehong Gao*, Deng-Ping Fan#, Linbo Jin, Ben Chen, Haoming Zhou, Minghui

250 Jan 08, 2023
O-CNN: Octree-based Convolutional Neural Networks for 3D Shape Analysis

O-CNN This repository contains the implementation of our papers related with O-CNN. The code is released under the MIT license. O-CNN: Octree-based Co

Microsoft 607 Dec 28, 2022
Hybrid CenterNet - Hybrid-supervised object detection / Weakly semi-supervised object detection

Hybrid-Supervised Object Detection System Object detection system trained by hybrid-supervision/weakly semi-supervision (HSOD/WSSOD): This project is

5 Dec 10, 2022
Codebase for arXiv preprint "NeRF++: Analyzing and Improving Neural Radiance Fields"

NeRF++ Codebase for arXiv preprint "NeRF++: Analyzing and Improving Neural Radiance Fields" Work with 360 capture of large-scale unbounded scenes. Sup

Kai Zhang 722 Dec 28, 2022
CPF: Learning a Contact Potential Field to Model the Hand-object Interaction

Contact Potential Field This repo contains model, demo, and test codes of our paper: CPF: Learning a Contact Potential Field to Model the Hand-object

Lixin YANG 99 Dec 26, 2022
Provided is code that demonstrates the training and evaluation of the work presented in the paper: "On the Detection of Digital Face Manipulation" published in CVPR 2020.

FFD Source Code Provided is code that demonstrates the training and evaluation of the work presented in the paper: "On the Detection of Digital Face M

88 Nov 22, 2022
The code of paper "Block Modeling-Guided Graph Convolutional Neural Networks".

Block Modeling-Guided Graph Convolutional Neural Networks This repository contains the demo code of the paper: Block Modeling-Guided Graph Convolution

22 Dec 08, 2022
PyTorch implementation of saliency map-aided GAN for Auto-demosaic+denosing

Saiency Map-aided GAN for RAW2RGB Mapping The PyTorch implementations and guideline for Saiency Map-aided GAN for RAW2RGB Mapping. 1 Implementations B

Yuzhi ZHAO 20 Oct 24, 2022
pybaum provides tools to work with pytrees which is a concept burrowed from JAX.

pybaum provides tools to work with pytrees which is a concept burrowed from JAX.

Open Source Economics 9 May 11, 2022
Sequence to Sequence (seq2seq) Recurrent Neural Network (RNN) for Time Series Forecasting

Sequence to Sequence (seq2seq) Recurrent Neural Network (RNN) for Time Series Forecasting Note: You can find here the accompanying seq2seq RNN forecas

Guillaume Chevalier 1k Dec 25, 2022
Anomaly Localization in Model Gradients Under Backdoor Attacks Against Federated Learning

Federated_Learning This repo provides a federated learning framework that allows to carry out backdoor attacks under varying conditions. This is a ker

Arçelik ARGE Açık Kaynak Yazılım Organizasyonu 0 Nov 30, 2021
Python Auto-ML Package for Tabular Datasets

Tabular-AutoML AutoML Package for tabular datasets Tabular dataset tuning is now hassle free! Run one liner command and get best tuning and processed

Sagnik Roy 18 Nov 20, 2022
a generic C++ library for image analysis

VIGRA Computer Vision Library Copyright 1998-2013 by Ullrich Koethe This file is part of the VIGRA computer vision library. You may use,

Ullrich Koethe 378 Dec 30, 2022
Enabling Lightweight Fine-tuning for Pre-trained Language Model Compression based on Matrix Product Operators

Enabling Lightweight Fine-tuning for Pre-trained Language Model Compression based on Matrix Product Operators This is our Pytorch implementation for t

RUCAIBox 12 Jul 22, 2022