Author | open – mmlab compile | Flin source | dead simple

Benchmarks and Model Zoo

The environment

hardware
  • 8 NVIDIA Tesla V100 GPUs
  • Intel Xeon 4114 CPU @ 2.20GHz
Software environment
  • Python 3.6/3.7
  • PyTorch 1.1
  • CUDA 9.0.176
  • CUDNN 7.0.4
  • NCCL 2.1.15

Mirror sites

We use AWS as the primary site for hosting Model Zoo and maintain the image on Ali Cloud. You can replace s3.ap-northeast-2.amazonaws.com/open-mmlab with…

Commonly used Settings

  • All FPN and RPN-C4 benchmarks were trained using 8 Gpus with a batch size of 16(2 images per GPU). The other C4 baselines were trained with 8 Gpus of batch size 8 (1 image per GPU).
  • All the models are herecoco_2017_trainOn training as well as incoco_2017_valThe test.
  • We use distributed training, and BN level statistics are fixed.
  • We use the same training schedule as Detectron. 1X represents 12 epochs, while 2X represents 24 Epochs, which is slightly less than the number of iterations of Detectron and negligible.
  • All of the PyTorch style pretraining trunks on ImageNet come from PyTorchmodel Zoo.
  • For a fair comparison with other code bases, we report GPU memorytorch.cuda.max_memory_allocated()Is the maximum value of all eight Gpus. Note that this value is usually less thannvidia-smiThe value displayed.
  • We report reasoning time as total time, including data loading, network forwarding, and post-processing.

The baseline

More models with different trunks will be added to the Model Zoo.

RPN
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) AR1000 Download
R-50-C4 caffe 1x 20.5 51.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 caffe 2x 2.2 0.17 20.3 52.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch 1x 20.1 50.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch 2x 20.0 51.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe 1x 3.3 0.253 16.9 58.2
R-50-FPN pytorch 1x 3.5 0.276 17.7 57.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 2x 57.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe 1x 5.2 0.379 13.9 59.4
R-101-FPN pytorch 1x 5.4 0.396 14.4 58.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 2x 59.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 1x 6.6 0.589 11.8 59.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 2x 59.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 1x 9.5 0.955 8.3 59.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 2x 60.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
Faster R-CNN
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP Download
R-50-C4 caffe 1x 9.5 34.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 caffe 2x 4.0 0.39 9.3 36.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch 1x 9.3 33.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch 2x 9.4 35.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe 1x 3.6 0.333 13.5 36.6
R-50-FPN pytorch 1x 3.8 0.353 13.6 36.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 2x 37.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe 1x 5.5 0.465 11.5 38.8
R-101-FPN pytorch 1x 5.7 0.474 11.9 38.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 2x 39.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 1x 6.9 0.672 10.3 40.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 2x 40.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 1x 9.8 1.040 7.3 41.3 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 2x 40.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
HRNetV2p-W18 pytorch 1x 36.1 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W18 pytorch 2x 38.3 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 1x 39.5 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 2x 40.6 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 1x 40.9 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 2x 41.5 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
Mask R-CNN
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP mask AP Download
R-50-C4 caffe 1x 8.1 35.9 31.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 caffe 2x 4.2 0.43 8.1 37.9 32.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch 1x 7.9 35.1 31.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch 2x 8.0 37.2 32.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe 1x 3.8 0.430 10.2 37.4 34.3
R-50-FPN pytorch 1x 3.9 0.453 10.6 37.3 34.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 2x 38.5 35.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe 1x 5.7 0.534 9.4 39.9 36.1
R-101-FPN pytorch 1x 5.8 0.571 9.5 39.4 35.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 2x 40.3 36.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 1x 7.1 0.759 8.3 41.1 37.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 2x 41.4 37.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 1x 10.0 1.102 6.5 42.1 38.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 2x 42.0 37.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
HRNetV2p-W18 pytorch 1x 37.3 34.2 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W18 pytorch 2x 39.2 35.7 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 1x 40.7 36.8 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 2x 41.7 37.5 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 1x 42.4 38.1 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 2x 42.9 38.3 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
Fast R-CNN (with pre-calculated proposals)
Backbone Style type Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP mask AP Download
R-50-C4 caffe Faster 1x 6.7 35.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 caffe Faster 2x 3.8 0.34 6.6 36.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch Faster 1x 6.3 34.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch Faster 2x 6.1 35.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe Faster 1x 3.3 0.242 18.4 36.6
R-50-FPN pytorch Faster 1x 3.5 0.250 16.5 35.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 caffe Mask 1x 8.1 35.9 31.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 caffe Mask 2x 4.2 0.43 8.1 37.9 32.9 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch Mask 1x 7.9 35.1 31.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-C4 pytorch Mask 2x 8.0 37.2 32.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch Faster 2x 37.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe Faster 1x 5.2 0.355 14.4 38.6
R-101-FPN pytorch Faster 1x 5.4 0.388 13.2 38.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch Faster 2x 38.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe Mask 1x 3.4 0.328 12.8 37.3 34.5
R-50-FPN pytorch Mask 1x 3.5 0.346 12.7 36.8 34.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch Mask 2x 37.9 34.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe Mask 1x 5.2 0.429 11.2 39.4 36.1
R-101-FPN pytorch Mask 1x 5.4 0.462 10.9 38.9 35.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch Mask 2x 39.9 36.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
RetinaNet
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP Download
R-50-FPN caffe 1x 3.4 0.285 12.5 35.8
R-50-FPN pytorch 1x 3.6 0.308 12.1 35.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 2x 36.4 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
R-101-FPN caffe 1x 5.3 0.410 10.4 37.8
R-101-FPN pytorch 1x 5.5 0.429 10.9 37.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 2x 38.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 1x 6.7 0.632 9.3 39.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 2x 39.3 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 1x 9.6 0.993 7.0 40.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 2x 39.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
Cascade R-CNN
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP Download
R-50-C4 caffe 1x 8.7 0.92 5.0 38.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe 1x 3.9 0.464 10.9 40.5
R-50-FPN pytorch 1x 4.1 0.455 11.9 40.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 20e 41.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe 1x 5.8 0.569 9.6 42.4
R-101-FPN pytorch 1x 6.0 0.584 10.3 42.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 20e 42.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 1x 7.2 0.770 8.9 43.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 20e 44.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 1x 10.0 1.133 6.7 44.5 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 20e 44.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
HRNetV2p-W18 pytorch 20e 41.2 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 20e 43.7 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 20e 44.6 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
Cascade Mask R-CNN
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP mask AP Download
R-50-C4 caffe 1x 9.1 0.99 4.5 39.3 32.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN caffe 1x 5.1 0.692 7.6 40.9 35.5
R-50-FPN pytorch 1x 5.3 0.683 7.4 41.2 35.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 20e 42.3 36.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN caffe 1x 7.0 0.803 7.2 43.1 37.2
R-101-FPN pytorch 1x 7.2 0.807 6.8 42.6 37.0 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 20e 43.3 37.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 1x 8.4 0.976 6.6 44.4 38.2 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 20e 44.7 38.6 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 1x 11.4 1.33 5.3 45.4 39.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 20e 45.7 39.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
HRNetV2p-W18 pytorch 20e 41.9 36.4 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 20e 44.5 38.5 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 20e 46.0 39.5 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)

S:

  • The schedule in 20E cascading (mask)R-CNN indicates lr reduction at 16th and 19th epochs, for a total reduction of 20 epochs.
Hybrid Mission Cascade (HTC)
Backbone Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP mask AP Download
R-50-FPN pytorch 1x 7.4 0.936 4.1 42.1 37.3 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-50-FPN pytorch 20e 43.2 38.1 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
R-101-FPN pytorch 20e 9.3 1.051 4.0 44.9 39.4 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-32x4d-FPN pytorch 20e 5.8 0.769 3.8 46.1 40.3 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
X-101-64x4d-FPN pytorch 20e 7.5 1.120 3.5 46.9 40.8 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
HRNetV2p-W18 pytorch 20e 43.1 37.9 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W32 pytorch 20e 45.3 39.6 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 20e 46.8 40.7 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)
HRNetV2p-W48 pytorch 28e 47.0 41.0 model(Open-mmlab.s3.ap-northeast-2.amazonaws.com/mmdetection…)

Note:

  • For more information and a more powerful model (50.7/43.9), see Mixed task cascading (github.com/open-mmlab/…) .
SSD
Backbone Size Style Lr schd Memory (GB) Training Time (S /iter) Minimum time (FPS) box AP Download
VGG16 300 caffe 120e 3.5 0.256 25.9/34.6 25.7 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)
VGG16 512 caffe 120e 7.6 0.412 20.7/25.4 29.3 model(S3.ap-northeast-2.amazonaws.com/open-mmlab/…)

Note:

  • cudnn.benchmarkSet toTrueUsed for SSD training and testing.
  • For batch size= 1 and Batch size= 8, the reasoning time is reported.
  • Because the model parameters and NMS,COCO and VOC speed are different.
Group normalization (GN)

For more information, see group normalization (github.com/open-mmlab/…

Weight standardization

For more information, see weight standardization (github.com/open-mmlab/…

Deformable convolution v2

For more information, see Deformable convolutional Networks (github.com/open-mmlab/…

CARAFE: Content-aware reorganization of functionality

For more information, please refer to CARAFE(github.com/open-mmlab/…

Instaboost

For more information, see Instaboost(github.com/open-mmlab/…

Libra R-CNN

For more information, see Libra R-CNN(github.com/open-mmlab/…

Guided Anchoring

For more information, see Guided Anchoring(github.com/open-mmlab/…

FCOS

For more information, see FCOS(github.com/open-mmlab/…

FoveaBox

For more information, see FoveaBox(github.com/open-mmlab/…

RepPoints

For more information, refer to RepPoints(github.com/open-mmlab/…

FreeAnchor

For more information, see FreeAnchor(github.com/open-mmlab/…

Grid R-CNN (plus)

For more information, see Grid R-CNN(github.com/open-mmlab/…

GHM

For more information, see GHM(github.com/open-mmlab/…

GCNet

For more information, please refer to GCNet(github.com/open-mmlab/…

HRNet

For more information, please refer to HRNet(github.com/open-mmlab/…

Mask Scoring R-CNN

For more information, please refer to Mask Scoring R-CNN(github.com/open-mmlab/…

Train from Scratch

For more information, see Rethinking ImageNet pre-training (github.com/open-mmlab/…

NAS-FPN

For more information, see NAS-FPN(github.com/open-mmlab/…

ATSS

For more information, please refer to ATSS(github.com/open-mmlab/…

Other data sets

We are also interested in PASCAL VOC(github.com/open-mmlab/… (github.com/open-mmlab/ MMDetection/blob/Master/Configs/Cityscapes) and WIDER FACE(github.com/open-mmlab/…

Comparison with Detectron and MasKRCNN-Benchmark

We combine MMDetection with Detectron(github.com/facebookres…) And maskrcnn – benchmark (github.com/facebookres…

In general, MMDetection has three advantages over Detectron.

  • Higher performance (especially for Mask AP)
  • Faster training speed
  • Efficient memory
performance

Detectron and MaskrCNN-Benchmark use Caffe-style ResNet as their backbone. We use caffe style (weights from (github.com/facebookres…) And PyTorch style (weights from official Model Zoo)ResNet trunk reports results, represented as PyTorch style results/Caffe style results.

We found that PyTorch style ResNet generally converges more slowly than Caffe style ResNet, resulting in slightly lower results at 1x progress, but higher final results at 2x progress.

type Lr schd Detectron maskrcnn-benchmark mmdetection
RPN 1x 57.2 57.1/58.2
2x 57.6 / –
Faster R-CNN 1x 36.7 36.8 36.4/36.6
2x 37.9 37.7 / –
Mask R-CNN 1x 37.7 & 33.9 37.8 & 34.2 37.3 & 34.2/37.4 & 34.3
2x 38.6 & 34.5 38.5&35.1 / –
Fast R-CNN 1x 36.4 35.8/36.6
2x 36.8 37.1 / –
Fast R-CNN (w/mask) 1x 37.3 & 33.7 36.8 & 34.1/37.3 & 34.5
2x 37.7 & 34.0 37.9&34.8 / –
Training speed

The training speed is measured in s/ ITER. The lower the better.

type Detectron (P1001) maskrcnn-benchmark (V100) mmdetection (V1002)
RPN 0.416 0.253
Faster R-CNN 0.544 0.353 0.333
Mask R-CNN 0.889 0.454 0.430
Fast R-CNN 0.285 0.242
Fast R-CNN (w/mask) 0.377 0.328
  • 1. Facebook’s Big Basin server (P100 / V100) is slightly faster than the server we use. Mmdetection also runs slightly faster on FB servers.

  • 2. For a fair comparison, we’ve listed Caffe’s results here.

Reasoning speed

Inference speed is measured in FPS (img/s) on a single GPU. The higher the better.

type Detectron (P100) maskrcnn-benchmark (V100) mmdetection (V100)
RPN 12.5 16.9
Faster R-CNN 10.3 7.9 13.5
Mask R-CNN 8.5 7.7 10.2
Fast R-CNN 12.5 18.4
Fast R-CNN (w/mask) 9.9 12.8
The training of memory
type Detectron maskrcnn-benchmark mmdetection
RPN 6.4 3.3
Faster R-CNN 7.2 4.4 3.6
Mask R-CNN 8.6 5.2 3.8
Fast R-CNN 6.0 3.3
Fast R-CNN (w/mask) 7.9 3.4

There is no doubt that maskrCNN benchmark and MMDetection are more efficient than Detectron for storage, but the main advantage is PyTorch itself. We also performed some memory optimizations to push it forward.

Note that Caffe2 and PyTorch have different apis to get memory usage through different implementations. For all code bases, the memory usage shown by NVIDIa-SMI is greater than the figure reported in the table above.

The original link: mmdetection. Readthedocs. IO/en/latest/M…

Welcome to panchuangai blog: panchuang.net/

Sklearn123.com/

Welcome to docs.panchuang.net/