[arXiv] [Slides] [Video]
Once-for-All: Train One Network and Specialize it for Efficient Deployment@inproceedings{
cai2020once,
title={Once for All: Train One Network and Specialize it for Efficient Deployment},
author={Han Cai and Chuang Gan and Tianzhe Wang and Zhekai Zhang and Song Han},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://arxiv.org/pdf/1908.09791.pdf}
}
[News] Once-for-All is available at PyTorch Hub now!
[News] Once-for-All (OFA) Network is adopted by SONY Neural Architecture Search Library.
[News] Once-for-All (OFA) Network is adopted by ADI MAX78000/MAX78002 Model Training and Synthesis Tool.
[News] Once-for-All (OFA) Network is adopted by Alibaba and ranked 1st in the open division of the MLPerf Inference Benchmark (Datacenter and Edge).
[News] First place in the CVPR 2020 Low-Power Computer Vision Challenge, CPU detection and FPGA track.
[News] OFA-ResNet50 is released.
[News] The hands-on tutorial of OFA is released!
[News] OFA is available via pip! Run pip install ofa
to install the whole OFA codebase.
[News] First place in the 4th Low-Power Computer Vision Challenge, both classification and detection track.
[News] First place in the 3rd Low-Power Computer Vision Challenge, DSP track at ICCV’19 using the Once-for-all Network.
Train once, specialize for many deployment scenarios
80% top1 ImageNet accuracy under mobile setting
Consistently outperforms MobileNetV3 on Diverse hardware platforms
[How to use]
OFA-ResNet50How to use / evaluate OFA Networks
Use
""" OFA Networks.
Example: ofa_network = ofa_net('ofa_mbv3_d234_e346_k357_w1.0', pretrained=True)
"""
from ofa.model_zoo import ofa_net
ofa_network = ofa_net(net_id, pretrained=True)
# Randomly sample sub-networks from OFA network
ofa_network.sample_active_subnet()
random_subnet = ofa_network.get_active_subnet(preserve_weight=True)
# Manually set the sub-network
ofa_network.set_active_subnet(ks=7, e=6, d=4)
manual_subnet = ofa_network.get_active_subnet(preserve_weight=True)
If the above scripts failed to download, you download it manually from Google Drive and put them under $HOME/.torch/ofa_nets/.
Evaluate
python eval_ofa_net.py --path 'Your path to imagenet' --net ofa_mbv3_d234_e346_k357_w1.0
OFA Network | Design Space | Resolution | Width Multiplier | Depth | Expand Ratio | kernel Size |
---|---|---|---|---|---|---|
ofa_resnet50 | ResNet50D | 128 - 224 | 0.65, 0.8, 1.0 | 0, 1, 2 | 0.2, 0.25, 0.35 | 3 |
ofa_mbv3_d234_e346_k357_w1.0 | MobileNetV3 | 128 - 224 | 1.0 | 2, 3, 4 | 3, 4, 6 | 3, 5, 7 |
ofa_mbv3_d234_e346_k357_w1.2 | MobileNetV3 | 160 - 224 | 1.2 | 2, 3, 4 | 3, 4, 6 | 3, 5, 7 |
ofa_proxyless_d234_e346_k357_w1.3 | ProxylessNAS | 128 - 224 | 1.3 | 2, 3, 4 | 3, 4, 6 | 3, 5, 7 |
How to use / evaluate OFA Specialized Networks
Use
""" OFA Specialized Networks.
Example: net, image_size = ofa_specialized('flops@[email protected]_finetune@75', pretrained=True)
"""
from ofa.model_zoo import ofa_specialized
net, image_size = ofa_specialized(net_id, pretrained=True)
If the above scripts failed to download, you download it manually from Google Drive and put them under $HOME/.torch/ofa_specialized/.
Evaluate
python eval_specialized_net.py --path 'Your path to imagent' --net flops@[email protected]_finetune@75
Model Name | Details | Top-1 (%) | Top-5 (%) | #Params | #MACs |
---|---|---|---|---|---|
ResNet50 Design Space | |||||
ofa-resnet50D-41 | [email protected][email protected] | 79.8 | 94.7 | 30.9M | 4.1B |
ofa-resnet50D-37 | [email protected][email protected] | 79.7 | 94.7 | 26.5M | 3.7B |
ofa-resnet50D-30 | [email protected][email protected] | 79.3 | 94.5 | 28.7M | 3.0B |
ofa-resnet50D-24 | [email protected][email protected] | 79.0 | 94.2 | 29.0M | 2.4B |
ofa-resnet50D-18 | [email protected][email protected] | 78.3 | 94.0 | 20.7M | 1.8B |
ofa-resnet50D-12 | [email protected][email protected]_finetune@25 | 77.1 | 93.3 | 19.3M | 1.2B |
ofa-resnet50D-09 | [email protected][email protected]_finetune@25 | 76.3 | 92.9 | 14.5M | 0.9B |
ofa-resnet50D-06 | [email protected][email protected]_finetune@25 | 75.0 | 92.1 | 9.6M | 0.6B |
FLOPs | |||||
ofa-595M | flops@[email protected]_finetune@75 | 80.0 | 94.9 | 9.1M | 595M |
ofa-482M | flops@[email protected]_finetune@75 | 79.6 | 94.8 | 9.1M | 482M |
ofa-389M | flops@[email protected]_finetune@75 | 79.1 | 94.5 | 8.4M | 389M |
LG G8 | |||||
ofa-lg-24 | LG-G8_lat@[email protected]_finetune@25 | 76.4 | 93.0 | 5.8M | 230M |
ofa-lg-16 | LG-G8_lat@[email protected]_finetune@25 | 74.7 | 92.0 | 5.8M | 151M |
ofa-lg-11 | LG-G8_lat@[email protected]_finetune@25 | 73.0 | 91.1 | 5.0M | 103M |
ofa-lg-8 | LG-G8_lat@[email protected]_finetune@25 | 71.1 | 89.7 | 4.1M | 74M |
Samsung S7 Edge | |||||
ofa-s7edge-88 | s7edge_lat@[email protected]_finetune@25 | 76.3 | 92.9 | 6.4M | 219M |
ofa-s7edge-58 | s7edge_lat@[email protected]_finetune@25 | 74.7 | 92.0 | 4.6M | 145M |
ofa-s7edge-41 | s7edge_lat@[email protected]_finetune@25 | 73.1 | 91.0 | 4.7M | 96M |
ofa-s7edge-29 | s7edge_lat@[email protected]_finetune@25 | 70.5 | 89.5 | 3.8M | 66M |
Samsung Note8 | |||||
ofa-note8-65 | note8_lat@[email protected]_finetune@25 | 76.1 | 92.7 | 5.3M | 220M |
ofa-note8-49 | note8_lat@[email protected]_finetune@25 | 74.9 | 92.1 | 6.0M | 164M |
ofa-note8-31 | note8_lat@[email protected]_finetune@25 | 72.8 | 90.8 | 4.6M | 101M |
ofa-note8-22 | note8_lat@[email protected]_finetune@25 | 70.4 | 89.3 | 4.3M | 67M |
Samsung Note10 | |||||
ofa-note10-64 | note10_lat@[email protected]_finetune@75 | 80.2 | 95.1 | 9.1M | 743M |
ofa-note10-50 | note10_lat@[email protected]_finetune@75 | 79.7 | 94.9 | 9.1M | 554M |
ofa-note10-41 | note10_lat@[email protected]_finetune@75 | 79.3 | 94.5 | 9.0M | 457M |
ofa-note10-30 | note10_lat@[email protected]_finetune@75 | 78.4 | 94.2 | 7.5M | 339M |
ofa-note10-22 | note10_lat@[email protected]_finetune@25 | 76.6 | 93.1 | 5.9M | 237M |
ofa-note10-16 | note10_lat@[email protected]_finetune@25 | 75.5 | 92.3 | 4.9M | 163M |
ofa-note10-11 | note10_lat@[email protected]_finetune@25 | 73.6 | 91.2 | 4.3M | 110M |
ofa-note10-08 | note10_lat@[email protected]_finetune@25 | 71.4 | 89.8 | 3.8M | 79M |
Google Pixel1 | |||||
ofa-pixel1-143 | pixel1_lat@[email protected]_finetune@75 | 80.1 | 95.0 | 9.2M | 642M |
ofa-pixel1-132 | pixel1_lat@[email protected]_finetune@75 | 79.8 | 94.9 | 9.2M | 593M |
ofa-pixel1-79 | pixel1_lat@[email protected]_finetune@75 | 78.7 | 94.2 | 8.2M | 356M |
ofa-pixel1-58 | pixel1_lat@[email protected]_finetune@75 | 76.9 | 93.3 | 5.8M | 230M |
ofa-pixel1-40 | pixel1_lat@[email protected]_finetune@25 | 74.9 | 92.1 | 6.0M | 162M |
ofa-pixel1-28 | pixel1_lat@[email protected]_finetune@25 | 73.3 | 91.0 | 5.2M | 109M |
ofa-pixel1-20 | pixel1_lat@[email protected]_finetune@25 | 71.4 | 89.8 | 4.3M | 77M |
Google Pixel2 | |||||
ofa-pixel2-62 | pixel2_lat@[email protected]_finetune@25 | 75.8 | 92.7 | 5.8M | 208M |
ofa-pixel2-50 | pixel2_lat@[email protected]_finetune@25 | 74.7 | 91.9 | 4.7M | 166M |
ofa-pixel2-35 | pixel2_lat@[email protected]_finetune@25 | 73.4 | 91.1 | 5.1M | 113M |
ofa-pixel2-25 | pixel2_lat@[email protected]_finetune@25 | 71.5 | 90.1 | 4.1M | 79M |
1080ti GPU (Batch Size 64) | |||||
ofa-1080ti-27 | 1080ti_gpu64@[email protected]_finetune@25 | 76.4 | 93.0 | 6.5M | 397M |
ofa-1080ti-22 | 1080ti_gpu64@[email protected]_finetune@25 | 75.3 | 92.4 | 5.2M | 313M |
ofa-1080ti-15 | 1080ti_gpu64@[email protected]_finetune@25 | 73.8 | 91.3 | 6.0M | 226M |
ofa-1080ti-12 | 1080ti_gpu64@[email protected]_finetune@25 | 72.6 | 90.9 | 5.9M | 165M |
V100 GPU (Batch Size 64) | |||||
ofa-v100-11 | v100_gpu64@[email protected]_finetune@25 | 76.1 | 92.7 | 6.2M | 352M |
ofa-v100-09 | v100_gpu64@[email protected]_finetune@25 | 75.3 | 92.4 | 5.2M | 313M |
ofa-v100-06 | v100_gpu64@[email protected]_finetune@25 | 73.0 | 91.1 | 4.9M | 179M |
ofa-v100-05 | v100_gpu64@[email protected]_finetune@25 | 71.6 | 90.3 | 5.2M | 141M |
Jetson TX2 GPU (Batch Size 16) | |||||
ofa-tx2-96 | tx2_gpu16@[email protected]_finetune@25 | 75.8 | 92.7 | 6.2M | 349M |
ofa-tx2-80 | tx2_gpu16@[email protected]_finetune@25 | 75.4 | 92.4 | 5.2M | 313M |
ofa-tx2-47 | tx2_gpu16@[email protected]_finetune@25 | 72.9 | 91.1 | 4.9M | 179M |
ofa-tx2-35 | tx2_gpu16@[email protected]_finetune@25 | 70.3 | 89.4 | 4.3M | 121M |
Intel Xeon CPU with MKL-DNN (Batch Size 1) | |||||
ofa-cpu-17 | cpu_lat@[email protected]_finetune@25 | 75.7 | 92.6 | 4.9M | 365M |
ofa-cpu-15 | cpu_lat@[email protected]_finetune@25 | 74.6 | 92.0 | 4.9M | 301M |
ofa-cpu-11 | cpu_lat@[email protected]_finetune@25 | 72.0 | 90.4 | 4.4M | 160M |
ofa-cpu-10 | cpu_lat@[email protected]_finetune@25 | 71.1 | 89.9 | 4.2M | 143M |
How to train OFA Networks
mpirun -np 32 -H <server1_ip>:8,<server2_ip>:8,<server3_ip>:8,<server4_ip>:8 \
-bind-to none -map-by slot \
-x NCCL_DEBUG=INFO -x LD_LIBRARY_PATH -x PATH \
python train_ofa_net.py
or
horovodrun -np 32 -H <server1_ip>:8,<server2_ip>:8,<server3_ip>:8,<server4_ip>:8 \
python train_ofa_net.py
Introduction Video
Hands-on Tutorial Video
Requirement
- Python 3.6+
- Pytorch 1.4.0+
- ImageNet Dataset
- Horovod
Related work on automated and efficient deep learning:
ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware (ICLR’19)
AutoML for Architecting Efficient and Specialized Neural Networks (IEEE Micro)
AMC: AutoML for Model Compression and Acceleration on Mobile Devices (ECCV’18)
HAQ: Hardware-Aware Automated Quantization (CVPR’19, oral)