PSPNet in Chainer

Last update: Dec 12, 2022

Overview

PSPNet

This is an unofficial implementation of Pyramid Scene Parsing Network (PSPNet) in Chainer.

Training

Requirement

Python 3.4.4+
- Chainer 3.0.0b1+
- ChainerMN master
- CuPy 2.0.0b1+
- ChainerCV 0.6.0+
- NumPy 1.12.0+
- tqdm 4.11.0+

pip install chainer --pre
pip install cupy --pre
pip install git+git://github.com/chainer/chainermn
pip install git+git://github.com/chainer/chainercv
pip install tqdm

Inference using converted weights

Requirement

Python 3.4.4+
- Chainer 3.0.0b1+
- ChainerCV 0.6.0+
- Matplotlib 2.0.0+
- CuPy 2.0.0b1+
- tqdm 4.11.0+

1. Run demo.py

Cityscapes

$ python demo.py -g 0 -m cityscapes -f aachen_000000_000019_leftImg8bit.png

Pascal VOC2012

$ python demo.py -g 0 -m voc2012 -f 2008_000005.jpg

ADE20K

$ python demo.py -g 0 -m ade20k -f ADE_val_00000001.jpg

FAQ

If you get RuntimeError: Invalid DISPLAY variable, how about specifying the matplotlib's backend by an environment variable?

$ MPLBACKEND=Agg python demo.py -g 0 -m cityscapes -f aachen_000000_000019_leftImg8bit.png

Convert weights by yourself

Caffe is NOT needed to convert .caffemodel to Chainer model. Use caffe_pb2.py.

Requirement

Python 3.4.4+
- protobuf 3.2.0+
- Chainer 3.0.0b1+
- NumPy 1.12.0+

1. Download the original weights

Please download the weights below from the author's repository:

pspnet50_ADE20K.caffemodel: GoogleDrive
pspnet101_VOC2012.caffemodel: GoogleDrive
pspnet101_cityscapes.caffemodel: GoogleDrive

and then put them into weights directory.

2. Convert weights

$ python convert.py

Reference

The original implementation by authors is: hszhao/PSPNet
The original paper is:
- Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, Jiaya Jia, "Pyramid Scene Parsing Network", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

Comments

Training failes with ModuleNotFoundError when using train_mn.py

Hi, I got following error when I tried to train PSP net with your train_mn.py How can I train my PSPNet model?

[email protected]:/yendo/oss/chainer-pspnet# python3 train_mn.py --result_dir result configs/cityscapes/pspnet.yml
Warning: using naive communicator because only naive supports CPU-only execution
==========================================
Num process (COMM_WORLD): 1
Using single_node communicator
Chainer version: 3.4.0
ChainerMN version: 1.2.0
cuda: True, cudnn: True
result_dir: result
Traceback (most recent call last):
  File "train_mn.py", line 504, in <module>
    trainer = get_trainer(args)
  File "train_mn.py", line 374, in get_trainer
    model = get_model_from_config(config, comm)
  File "train_mn.py", line 239, in get_model_from_config
    loss.module, loss.name, loss.args, comm)
  File "train_mn.py", line 219, in get_model
    mod = import_module(loss_module)
  File "/root/.pyenv/versions/anaconda3-5.0.1/lib/python3.6/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 994, in _gcd_import
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load
  File "<frozen importlib._bootstrap>", line 941, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 994, in _gcd_import
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load
  File "<frozen importlib._bootstrap>", line 953, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'loss'

opened by jo7ueb 0

Training Fails with IndexError when using train.py

Hi, I got following error when I tried to train PSP net with your train.py How can I train my PSPNet model?

[email protected]:/yendo/oss/chainer-pspnet# python3 train.py --gpu --result_dir result configs/cityscapes/pspnet.yml
==========================================
Chainer version: 3.4.0
CuPy version: 2.4.0
Traceback (most recent call last):
  File "train.py", line 483, in <module>
    trainer = get_trainer(args)
  File "train.py", line 339, in get_trainer
    chainer.cuda.available, chainer.cuda.cudnn_enabled, ))
IndexError: tuple index out of range

opened by jo7ueb 0

could you actually train a new model?

Hi, I am currently trying to train the cityscapes dataset with your code, but the result is miserable: still 0.5263158 (=1/19) class accuracy after 120 epochs. Apparently, the loss of training data is converged correctly, so it seems like a perfect over fitting. Since I used the same settings as yours, i am wondering how you managed to reproduce the results(maybe i need less learning rate?). thanks in advance!

opened by suzukikbp 0

Releases(ChainerCV_PSPNet)

ChainerCV_PSPNet(Aug 14, 2017)

Source code(tar.gz)
Source code(zip)
pspnet101_cityscapes_713_reference.npz(250.74 MB)
pspnet101_VOC2012_473_reference.npz(250.82 MB)
pspnet50_ADE20K_473_reference.npz(185.12 MB)
PSPNet_reference_weights(Aug 11, 2017)

Source code(tar.gz)
Source code(zip)
pspnet101_cityscapes_713_reference.chainer(233.76 MB)
pspnet101_VOC2012_473_reference.chainer(233.80 MB)
pspnet50_ADE20K_473_reference.chainer(166.00 MB)

Owner

Shunta Saito

Ph.D in Engineering, Researcher at Preferred Networks, Inc.

GitHub Repository

Tensorboard for pytorch (and chainer, mxnet, numpy, ...)

tensorboardX Write TensorBoard events with simple function call. The current release (v2.3) is tested on anaconda3, with PyTorch 1.8.1 / torchvision 0

7.5k Dec 28, 2022

LexGLUE: A Benchmark Dataset for Legal Language Understanding in English

LexGLUE: A Benchmark Dataset for Legal Language Understanding in English ⚖️ 🏆 🧑‍🎓 👩‍⚖️ Dataset Summary Inspired by the recent widespread use of th

95 Dec 08, 2022

Visualize Camera's Pose Using Extrinsic Parameter by Plotting Pyramid Model on 3D Space

extrinsic2pyramid Visualize Camera's Pose Using Extrinsic Parameter by Plotting Pyramid Model on 3D Space Intro A very simple and straightforward modu

106 Dec 28, 2022

Building blocks for uncertainty-aware cycle consistency presented at NeurIPS'21.

UncertaintyAwareCycleConsistency This repository provides the building blocks and the API for the work presented in the NeurIPS'21 paper Robustness vi

19 Dec 12, 2022

A Python library that enables ML teams to share, load, and transform data in a collaborative, flexible, and efficient way :chestnut:

Squirrel Core Share, load, and transform data in a collaborative, flexible, and efficient way What is Squirrel? Squirrel is a Python library that enab

249 Dec 07, 2022

Rule-based Customer Segmentation

Rule-based Customer Segmentation Business Problem A game company wants to create level-based new customer definitions (personas) by using some feature

2 Jan 03, 2022

Session-based Recommendation, CoHHN, price preferences, interest preferences, Heterogeneous Hypergraph, Co-guided Learning, SIGIR2022

This is our implementation for the paper: Price DOES Matter! Modeling Price and Interest Preferences in Session-based Recommendation Xiaokun Zhang, Bo

27 Dec 02, 2022

Official Pytorch implementation of "Unbiased Classification Through Bias-Contrastive and Bias-Balanced Learning (NeurIPS 2021)

Unbiased Classification Through Bias-Contrastive and Bias-Balanced Learning (NeurIPS 2021) Official Pytorch implementation of Unbiased Classification

17 Jan 01, 2023

PSPNet in Chainer

Related tags

Overview

PSPNet

Training

Requirement

Inference using converted weights

Requirement

1. Run demo.py

Cityscapes

Pascal VOC2012

ADE20K

FAQ

Convert weights by yourself

Requirement

1. Download the original weights

2. Convert weights

Reference

You might also like...

Comments

Training failes with ModuleNotFoundError when using train_mn.py

Training Fails with IndexError when using train.py

could you actually train a new model?

Releases(ChainerCV_PSPNet)

ChainerCV_PSPNet(Aug 14, 2017)

PSPNet_reference_weights(Aug 11, 2017)

Owner

Shunta Saito

Tensorboard for pytorch (and chainer, mxnet, numpy, ...)

LexGLUE: A Benchmark Dataset for Legal Language Understanding in English

Visualize Camera's Pose Using Extrinsic Parameter by Plotting Pyramid Model on 3D Space

Building blocks for uncertainty-aware cycle consistency presented at NeurIPS'21.

A Python library that enables ML teams to share, load, and transform data in a collaborative, flexible, and efficient way :chestnut:

Rule-based Customer Segmentation

Session-based Recommendation, CoHHN, price preferences, interest preferences, Heterogeneous Hypergraph, Co-guided Learning, SIGIR2022

Official Pytorch implementation of "Unbiased Classification Through Bias-Contrastive and Bias-Balanced Learning (NeurIPS 2021)

Pytorch implementation of our method for regularizing nerual radiance fields for few-shot neural volume rendering.

Set of models for classifcation of 3D volumes

High-Resolution Image Synthesis with Latent Diffusion Models

Official Repository for our ICCV2021 paper: Continual Learning on Noisy Data Streams via Self-Purified Replay

Open source implementation of "A Self-Supervised Descriptor for Image Copy Detection" (SSCD).

Educational API for 3D Vision using pose to control carton.

So-ViT: Mind Visual Tokens for Vision Transformer

LinkNet - This repository contains our Torch7 implementation of the network developed by us at e-Lab.

A general-purpose encoder-decoder framework for Tensorflow

Generative Models for Graph-Based Protein Design

RANZCR-CLiP 7th Place Solution

face2comics by Sxela (Alex Spirin) - face2comics datasets