Add ResNeSt 50, 101, 200, 269 MXNet Gluon Model Weights

ResNeSt 50, 101, 200, 269 model & detectron2 backbones

Change All Models Using Wasabi Url

Update PyTorch S3 Url to Wasabi to save money.

ResNeSt 0.0.3

Detectron2-ResNeSt

See details in Detectron2-ResNeSt repo

Object Detection on MS-COCO validation set

Method	Backbone	mAP%
Faster R-CNN	ResNet-50	39.25
	ResNet-101	41.37
	ResNeSt-50 (ours)	42.33
	ResNeSt-101 (ours)	44.72
Cascade R-CNN	ResNet-50	42.52
	ResNet-101	44.03
	ResNeSt-50 (ours)	45.41
	ResNeSt-101 (ours)	47.50
	ResNeSt-200 (ours)	49.03

Instance Segmentation

Method	Backbone	bbox	mask
Mask R-CNN	ResNet-50	39.97	36.05
	ResNet-101	41.78	37.51
	ResNeSt-50 (ours)	42.81	38.14
	ResNeSt-101 (ours)	45.75	40.65
Cascade R-CNN	ResNet-50	43.06	37.19
	ResNet-101	44.79	38.52
	ResNeSt-50 (ours)	46.19	39.55
	ResNeSt-101 (ours)	48.30	41.56

Semantic Segmentation with New SoTA on ADE20K

Semantic Segmentation

• PyTorch models and training: Please visit PyTorch Encoding Toolkit.
• Training with Gluon: Please visit GluonCV Toolkit.

Results on ADE20K

Method	Backbone	pixAcc%	mIoU%
Deeplab-V3	ResNet-50	80.39	42.1
	ResNet-101	81.11	44.14
	ResNeSt-50 (ours)	81.17	45.12
	ResNeSt-101 (ours)	82.07	46.91
	ResNeSt-269 (ours)	82.62	47.60

Ablation Study Models

	setting	#P	GFLOPs	PyTorch	Gluon
ResNeSt-50-fast	1s1x64d	26.3M	4.34	80.33	80.35
ResNeSt-50-fast	2s1x64d	27.5M	4.34	80.53	80.65
ResNeSt-50-fast	4s1x64d	31.9M	4.35	80.76	80.90
ResNeSt-50-fast	1s2x40d	25.9M	4.38	80.59	80.72
ResNeSt-50-fast	2s2x40d	26.9M	4.38	80.61	80.84
ResNeSt-50-fast	4s2x40d	30.4M	4.41	81.14	81.17
ResNeSt-50-fast	1s4x24d	25.7M	4.42	80.99	80.97

ImageNet Training with MXNet Gluon

Install MXNet with Horovod

# assuming you have CUDA 10.0 on your machine
pip install mxnet-cu100
HOROVOD_GPU_ALLREDUCE=NCCL pip install -v --no-cache-dir horovod
pip install --no-cache mpi4py

Prepare ImageNet recordio data format

• Unfortunately ,this is required for training using MXNet Gluon. Please follow the GluonCV tutorial to prepare the data.

• Copy the data into ramdisk (optional):

  d ~/
  udo mkdir -p /media/ramdisk
  udo mount -t tmpfs -o size=200G tmpfs /media/ramdisk
  p -r /home/ubuntu/data/ILSVRC2012/ /media/ramdisk
  

Training command

Using ResNeSt-50 as the target model:

horovodrun -np 64 --hostfile hosts python train.py \
--rec-train /media/ramdisk/ILSVRC2012/train.rec \
--rec-val /media/ramdisk/ILSVRC2012/val.rec \
--model resnest50 --lr 0.05 --num-epochs 270 --batch-size 128 \
--use-rec --dtype float32 --warmup-epochs 5 --last-gamma --no-wd \
--label-smoothing --mixup --save-dir params_ resnest50 \
--log-interval 50 --eval-frequency 5 --auto_aug --input-size 224

Verify pretrained model

python verify.py --model resnest50 --crop-size 224 --resume params_ resnest50/imagenet-resnest50-269.params

Pretrained Models

	crop size	PyTorch	Gluon
ResNeSt-50	224	81.03	81.04
ResNeSt-101	256	82.83	82.81
ResNeSt-200	320	83.84	83.88
ResNeSt-269	416	84.54	84.53

PyTorch Models

• Load using Torch Hub

import torch
# get list of models
torch.hub.list('zhanghang1989/ResNeSt', force_reload=True)

# load pretrained models, using ResNeSt-50 as an example
net = torch.hub.load('zhanghang1989/ResNeSt', 'resnest50', pretrained=True)

• Load using python package

# using ResNeSt-50 as an example
from resnest.torch import resnest50
net = resnest50(pretrained=True)

Gluon Models

• Load pretrained model:

# using ResNeSt-50 as an example
from resnest.gluon import resnest50
net = resnest50(pretrained=True)