Reproduce ResNet-v2(Identity Mappings in Deep Residual Networks) with MXNet

Related tags

Deep LearningResNet
Overview

Reproduce ResNet-v2 using MXNet

Requirements

  • Install MXNet on a machine with CUDA GPU, and it's better also installed with cuDNN v5
  • Please fix the randomness if you want to train your own model and using this pull request

Trained models

The trained ResNet models achieve better error rates than the original ResNet-v1 models.

ImageNet 1K

Imagenet 1000 class dataset with 1.2 million images.

single center crop (224x224) validation error rate(%)

Network Top-1 error Top-5 error Traind Model
ResNet-18 30.48 10.92 data.dmlc.ml
ResNet-34 27.20 8.86 data.dmlc.ml
ResNet-50 24.39 7.24 data.dmlc.ml
ResNet-101 22.68 6.58 data.dmlc.ml
ResNet-152 22.25 6.42 data.dmlc.ml
ResNet-200 22.14 6.16 data.dmlc.ml

ImageNet 11K:

Full imagenet dataset: fall11_whole.tar from http://www.image-net.org/download-images.

We removed classes with less than 500 images. The filtered dataset contains 11221 classes and 12.4 millions images. We randomly pick 50 images from each class as the validation set. The split is available at http://data.dmlc.ml/mxnet/models/imagenet-11k/

Network Top-1 error Top-5 error Traind Model
ResNet-200 58.4 28.8

cifar10: single crop validation error rate(%):

Network top-1
ResNet-164 4.68

Training Curve

The following curve is ResNet-v2 trainined on imagenet-1k, all the training detail you can found here, which include gpu information, lr schedular, batch-size etc, and you can also see the training speed with the corresponding logs.

you can get the curve by run:
cd log && python plot_curve.py --logs=resnet-18.log,resnet-34.log,resnet-50.log,resnet-101.log,resnet-152.log,resnet-200.log

How to Train

imagenet

first you should prepare the train.lst and val.lst, you can generate this list files by yourself(please ref.make-the-image-list, and do not forget to shuffle the list files!), or just download the provided version from here.

then you can create the *.rec file, i recommend use this cmd parameters:

$im2rec_path train.lst train/ data/imagenet/train_480_q90.rec resize=480 quality=90

set resize=480 and quality=90(quality=100 will be best i think:)) here may use more disk memory(about ~103G), but this is very useful with scale augmentation during training[1][2], and can help reproducing a good result.

because you are training imagenet , so we should set data-type = imagenet, then the training cmd is like this(here i use 6 gpus for training):

python -u train_resnet.py --data-dir data/imagenet \
--data-type imagenet --depth 50 --batch-size 256  --gpus=0,1,2,3,4,5

change depth to different number to support different model, currently support ResNet-18, ResNet-34, ResNet-50, ResNet-101, ResNet-152, ResNet-200.

cifar10

same as above, first you should use im2rec to create the .rec file, then training with cmd like this:

python -u train_resnet.py --data-dir data/cifar10 --data-type cifar10 \
  --depth 164 --batch-size 128 --num-examples 50000 --gpus=0,1

change depth when training different model, only support(depth-2)%9==0, such as RestNet-110, ResNet-164, ResNet-1001...

retrain

When training large dataset(like imagenet), it's better for us to change learning rate manually, or the training is killed by some other reasons, so retrain is very important. the code here support retrain, suppose you want to retrain your resnet-50 model from epoch 70 and want to change lr=0.0005, wd=0.001, batch-size=256 using 8gpu, then you can try this cmd:

python -u train_resnet.py --data-dir data/imagenet --data-type imagenet --depth 50 --batch-size 256 \
--gpus=0,1,2,3,4,5,6,7 --model-load-epoch=70 --lr 0.0005 --wd 0.001 --retrain

Notes

  • it's better training the model in imagenet with epoch > 110, because this will lead better result.
  • when epoch is about 95, cancel the scale/color/aspect augmentation during training, this can be done by only comment out 6 lines of the code, like this:
train = mx.io.ImageRecordIter(
        # path_imgrec         = os.path.join(args.data_dir, "train_480_q90.rec"),
        path_imgrec         = os.path.join(args.data_dir, "train_256_q90.rec"),
        label_width         = 1,
        data_name           = 'data',
        label_name          = 'softmax_label',
        data_shape          = (3, 32, 32) if args.data_type=="cifar10" else (3, 224, 224),
        batch_size          = args.batch_size,
        pad                 = 4 if args.data_type == "cifar10" else 0,
        fill_value          = 127,  # only used when pad is valid
        rand_crop           = True,
        # max_random_scale    = 1.0 if args.data_type == "cifar10" else 1.0,  # 480
        # min_random_scale    = 1.0 if args.data_type == "cifar10" else 0.533,  # 256.0/480.0
        # max_aspect_ratio    = 0 if args.data_type == "cifar10" else 0.25,
        # random_h            = 0 if args.data_type == "cifar10" else 36,  # 0.4*90
        # random_s            = 0 if args.data_type == "cifar10" else 50,  # 0.4*127
        # random_l            = 0 if args.data_type == "cifar10" else 50,  # 0.4*127
        rand_mirror         = True,
        shuffle             = True,
        num_parts           = kv.num_workers,
        part_index          = kv.rank)

but you should prepare one train_256_q90.rec using im2rec like:

$im2rec_path train.lst train/ data/imagenet/train_256_q90.rec resize=256 quality=90

cancel this scale/color/aspect augmentation can be done easily by using --aug-level=1 in your cmd.

  • it's better for running longer than 30 epoch before first decrease the lr(such as 60), so you may decide the epoch number by observe the val-acc curve, and set lr with retrain.

Training ResNet-200 by only one gpu with 'dark knowledge' of mxnet

you can training ResNet-200 or even ResNet-1000 on imaget with only one gpu! for example, we can train ResNet-200 with batch-size=128 on one gpu(=12G), or if your gpu memory is less than 12G, you should decrease the batch-size by a little. here is the way of how to using 'dark knowledge' of mxnet:

when turn on memonger, the trainning speed will be about 25% slower, but we can training more depth network, have fun!

ResNet-v2 vs ResNet-v1

Does ResNet-v2 always achieve better result than ResNet-v1 on imagnet? The answer is NO, ResNet-v2 has no advantage or even has disadvantage than ResNet-v1 when depth<152, we can get the following result from paper[2].(why?)

ImageNet: single center crop validation error rate(%)

Network crop-size top-1 top-5
ResNet-101-v1 224x224 23.6 7.1
ResNet-101-v2 224x224 24.6 7.5
ResNet-152-v1 320x320 21.3 5.5
ResNet-152-v2 320x320 21.1 5.5

we can see that:

  • when depth=101, ResNet-v2 is 1% worse than ResNet-v1 on top-1 and 0.4% worse on top-5.
  • when depth=152, ResNet-v2 is only 0.2% better than ResNet-v1 on top-1 and owns the same performance on top-5 even when crop-size=320x320.

How to use Trained Models

we can use the pre-trained model to classify one input image, the step is easy:

  • download the pre-trained model form data.dml.ml and put it into the predict directory.
  • cd predict and run python -u predict.py --img test.jpg --prefix resnet-50 --gpu 0, this means you want to recgnition test.jpg using model resnet-50-0000.params and gpu 0, then it will output the classification result.

Reference

[1] Kaiming He, et al. "Deep Residual Learning for Image Recognition." arXiv arXiv:1512.03385 (2015).
[2] Kaiming He, et al. "Identity Mappings in Deep Residual Networks" arXiv:1603.05027 (2016).
[3] caffe official training code and model, https://github.com/KaimingHe/deep-residual-networks
[4] torch training code and model provided by facebook, https://github.com/facebook/fb.resnet.torch
[5] MXNet resnet-v1 cifar10 examples,https://github.com/dmlc/mxnet/blob/master/example/image-classification/train_cifar10_resnet.py

Owner
Wei Wu
Wei Wu
Object detection and instance segmentation toolkit based on PaddlePaddle.

Object detection and instance segmentation toolkit based on PaddlePaddle.

9.3k Jan 02, 2023
Official implementation of "Dynamic Anchor Learning for Arbitrary-Oriented Object Detection" (AAAI2021).

DAL This project hosts the official implementation for our AAAI 2021 paper: Dynamic Anchor Learning for Arbitrary-Oriented Object Detection [arxiv] [c

ming71 215 Nov 28, 2022
CL-Gym: Full-Featured PyTorch Library for Continual Learning

CL-Gym: Full-Featured PyTorch Library for Continual Learning CL-Gym is a small yet very flexible library for continual learning research and developme

Iman Mirzadeh 36 Dec 25, 2022
ECLARE: Extreme Classification with Label Graph Correlations

ECLARE ECLARE: Extreme Classification with Label Graph Correlations @InProceedings{Mittal21b, author = "Mittal, A. and Sachdeva, N. and Agrawal

Extreme Classification 35 Nov 06, 2022
Powerful and efficient Computer Vision Annotation Tool (CVAT)

Computer Vision Annotation Tool (CVAT) CVAT is free, online, interactive video and image annotation tool for computer vision. It is being used by our

OpenVINO Toolkit 8.6k Jan 01, 2023
pcnaDeep integrates cutting-edge detection techniques with tracking and cell cycle resolving models.

pcnaDeep: a deep-learning based single-cell cycle profiler with PCNA signal Welcome! pcnaDeep integrates cutting-edge detection techniques with tracki

ChanLab 8 Oct 18, 2022
The Submission for SIMMC 2.0 Challenge 2021

The Submission for SIMMC 2.0 Challenge 2021 challenge website Requirements python 3.8.8 pytorch 1.8.1 transformers 4.8.2 apex for multi-gpu nltk Prepr

5 Jul 26, 2022
A Benchmark For Measuring Systematic Generalization of Multi-Hierarchical Reasoning

Orchard Dataset This repository contains the code used for generating the Orchard Dataset, as seen in the Multi-Hierarchical Reasoning in Sequences: S

Bill Pung 1 Jun 05, 2022
Prototype for Baby Action Detection and Classification

Baby Action Detection Table of Contents About Install Run Predictions Demo About An attempt to harness the power of Deep Learning to come up with a so

Shreyas K 30 Dec 16, 2022
A hifiasm fork for metagenome assembly using Hifi reads.

hifiasm_meta - de novo metagenome assembler, based on hifiasm, a haplotype-resolved de novo assembler for PacBio Hifi reads.

44 Jul 10, 2022
CVPR 2021: "Generating Diverse Structure for Image Inpainting With Hierarchical VQ-VAE"

Diverse Structure Inpainting ArXiv | Papar | Supplementary Material | BibTex This repository is for the CVPR 2021 paper, "Generating Diverse Structure

152 Nov 04, 2022
A large-scale face dataset for face parsing, recognition, generation and editing.

CelebAMask-HQ [Paper] [Demo] CelebAMask-HQ is a large-scale face image dataset that has 30,000 high-resolution face images selected from the CelebA da

switchnorm 1.7k Dec 26, 2022
BabelCalib: A Universal Approach to Calibrating Central Cameras. In ICCV (2021)

BabelCalib: A Universal Approach to Calibrating Central Cameras This repository contains the MATLAB implementation of the BabelCalib calibration frame

Yaroslava Lochman 55 Dec 30, 2022
Differentiable architecture search for convolutional and recurrent networks

Differentiable Architecture Search Code accompanying the paper DARTS: Differentiable Architecture Search Hanxiao Liu, Karen Simonyan, Yiming Yang. arX

Hanxiao Liu 3.7k Jan 09, 2023
A python library to build Model Trees with Linear Models at the leaves.

A python library to build Model Trees with Linear Models at the leaves.

Marco Cerliani 212 Dec 30, 2022
The code of NeurIPS 2021 paper "Scalable Rule-Based Representation Learning for Interpretable Classification".

Rule-based Representation Learner This is a PyTorch implementation of Rule-based Representation Learner (RRL) as described in NeurIPS 2021 paper: Scal

Zhuo Wang 53 Dec 17, 2022
FAST-RIR: FAST NEURAL DIFFUSE ROOM IMPULSE RESPONSE GENERATOR

This is the official implementation of our neural-network-based fast diffuse room impulse response generator (FAST-RIR) for generating room impulse responses (RIRs) for a given acoustic environment.

Anton Jeran Ratnarajah 89 Dec 22, 2022
3D-aware GANs based on NeRF (arXiv).

CIPS-3D This repository will contain the code of the paper, CIPS-3D: A 3D-Aware Generator of GANs Based on Conditionally-Independent Pixel Synthesis.

Peterou 563 Dec 31, 2022
Code for Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions

EMS-COLS-recourse Initial Code for Low-Cost Algorithmic Recourse for Users With Uncertain Cost Functions Folder structure: data folder contains raw an

Prateek Yadav 1 Nov 25, 2022
3D HourGlass Networks for Human Pose Estimation Through Videos

3D-HourGlass-Network 3D CNN Based Hourglass Network for Human Pose Estimation (3D Human Pose) from videos. This was my summer'18 research project. Dis

Naman Jain 51 Jan 02, 2023