Official PyTorch Implementation
Tal Ridnik, Hussam Lawen, Asaf Noy, Itamar Friedman
DAMO Academy, Alibaba Group
Abstract
Many deep learning models, developed in recent years, reach higher ImageNet accuracy than ResNet50, with fewer or comparable FLOPS count. While FLOPs are often seen as a proxy for network efficiency, when measuring actual GPU training and inference throughput, vanilla ResNet50 is usually significantly faster than its recent competitors, offering better throughput-accuracy trade-off. In this work, we introduce a series of architecture modifications that aim to boost neural networks' accuracy, while retaining their GPU training and inference efficiency. We first demonstrate and discuss the bottlenecks induced by FLOPs-optimizations. We then suggest alternative designs that better utilize GPU structure and assets. Finally, we introduce a new family of GPU-dedicated models, called TResNet, which achieve better accuracy and efficiency than previous ConvNets. Using a TResNet model, with similar GPU throughput to ResNet50, we reach 80.7% top-1 accuracy on ImageNet. Our TResNet models also transfer well and achieve state-of-the-art accuracy on competitive datasets such as Stanford cars (96.0%), CIFAR-10 (99.0%), CIFAR-100 (91.5%) and Oxford-Flowers (99.1%)
TResNet models accuracy and GPU throughput on ImageNet, compared to ResNet50. All measurements were done on Nvidia V100 GPU, with mixed precision. All models are trained on input resolution of 224.
Models | Top Training Speed (img/sec) |
Top Inference Speed (img/sec) |
Max Train Batch Size | Top-1 Acc. |
---|---|---|---|---|
ResNet50 | 805 | 2830 | 288 | 79.0 |
EfficientNetB1 | 440 | 2740 | 196 | 79.2 |
TResNet-M | 730 | 2930 | 512 | 80.7 |
TResNet-L | 345 | 1390 | 316 | 81.4 |
TResNet-XL | 250 | 1060 | 240 | 82.0 |
Comparison of ResNet50 to top modern networks, with similar top-1 ImageNet accuracy. All measurements were done on Nvidia V100 GPU with mixed precision. For gaining optimal speeds, training and inference were measured on 90% of maximal possible batch size. Except TResNet-M, all the models' ImageNet scores were taken from the public repository, which specialized in providing top implementations for modern networks. Except EfficientNet-B1, which has input resolution of 240, all other models have input resolution of 224.
Model | Top Training Speed (img/sec) |
Top Inference Speed (img/sec) |
Top-1 Acc. | Flops[G] |
---|---|---|---|---|
ResNet50 | 805 | 2830 | 79.0 | 4.1 |
ResNet50-D | 600 | 2670 | 79.3 | 4.4 |
ResNeXt50 | 490 | 1940 | 78.5 | 4.3 |
EfficientNetB1 | 440 | 2740 | 79.2 | 0.6 |
SEResNeXt50 | 400 | 1770 | 79.0 | 4.3 |
MixNet-L | 400 | 1400 | 79.0 | 0.5 |
TResNet-M | 730 | 2930 | 80.7 | 5.5 |
![]() |
![]() |
Comparison of TResNet to state-of-the-art models on transfer learning datasets (only ImageNet-based transfer learning results). Models inference speed is measured on a mixed precision V100 GPU. Since no official implementation of Gpipe was provided, its inference speed is unknown
Dataset | Model |
Top-1
Acc. |
Speed
img/sec |
Input |
CIFAR-10 | Gpipe | 99.0 | - | 480 |
TResNet-XL | 99.0 | 1060 | 224 | |
CIFAR-100 | EfficientNet-B7 | 91.7 | 70 | 600 |
TResNet-XL | 91.5 | 1060 | 224 | |
Stanford Cars | EfficientNet-B7 | 94.7 | 70 | 600 |
TResNet-L | 96.0 | 500 | 368 | |
Oxford-Flowers | EfficientNet-B7 | 98.8 | 70 | 600 |
TResNet-L | 99.1 | 500 | 368 |
We provide code for reproducing the validation top-1 score of TResNet models on ImageNet (input resolution 224). First, download pretrained models from here.
Then, run the infer.py script. For example, for tresnet_m run:
python -m infer.py \
--val_dir=/path/to/imagenet_val_folder \
--model_path=/model/path/to/tresnet_m.pth \
--model_name=tresnet_m
@misc{ridnik2020tresnet,
title={TResNet: High Performance GPU-Dedicated Architecture},
author={Tal Ridnik and Hussam Lawen and Asaf Noy and Itamar Friedman},
year={2020},
eprint={2003.13630},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
Feel free to contact me if there are any questions or issues (Tal Ridnik, [email protected]).