Skip to content

Add the config file of mixed precision quantization for ResNet50 #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
lyg95 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add the config file of mixed precision quantization for ResNet50 #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 41 additions & 7 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
# Model Optimizer

[![Build Status](https://dev.azure.com/Adlik/GitHub/_apis/build/status/Adlik.model_optimizer?branchName=main)](https://dev.azure.com/Adlik/GitHub/_build/results?buildId=3472&view=results)
[![Build Status](https://dev.azure.com/Adlik/GitHub/_apis/build/status/Adlik.model_optimizer?branchName=main)](https://dev.azure.com/Adlik/GitHub/_build/latest?definitionId=2&branchName=main)
[![Bors enabled](https://bors.tech/images/badge_small.svg)](https://app.bors.tech/repositories/65566)
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)

Expand All @@ -22,9 +22,17 @@ of the models. Note that activations use the same observer as weights unless oth
| resnet50 | 76.13 | 75.580 | 75.612 | 75.99 |
| mobilenetv2 | 71.878 | 70.730(act=percentile) | 70.816 | 71.11 |

For quantization, we explored low-bit model quantization, where weights are quantized to 3 bits and activations are
quantized to 4 bits, using the lsq algorithm to quantize resnet50. Compared with the FP32 model, the quantized resnet50
has no accuracy loss and has higher accuracy than the original model. Our quantized resnet50 model achieves 77.34%
accuracy on the ImageNet dataset. Here we supply the quantized model in
[model_zoo](https://github.com/Adlik/model_zoo) for testing. Besides, we submit our quantized model to
[paperswithcode](https://paperswithcode.com/sota/quantization-on-imagenet?tag_filter=447), this is the current
state-of-art result in low-bit quantization.

For AutoSlim, we give the pruning effect of the resnet50 model on the ImageNet dataset.

| FLOPs(G) |Params(M) |Size(MB)| Top-1| Acc |Input Size|
| Model | FLOPs(G) |Params(M) |Size(MB)| Top-1 Acc |Input Size|
|:---:| :---: | :---: | :---: | :---: | :---: |
|ResNet5 |4.12 |25.56 |98 |77.39% |224|
|ResNet-50 0.75× |2.35 |14.77 |57 |75.87% |224|
Expand All @@ -37,8 +45,8 @@ For AutoSlim, we give the pruning effect of the resnet50 model on the ImageNet d

The following table shows the effect of AutoSlim on YOLOv5m backbone pruning.

| FLOPs(G) |Params(M) |Size(MB)|mAPval 0.5:0.95| Input Size|
| :---: | :---: | :---: | :---: | :---: |
| Model | FLOPs(G) |Params(M) |Size(MB)|mAPval 0.5:0.95| Input Size|
| :---: | :---: | :---: | :---: | :---: | :---: |
|YOLOv5m |24.5 |21.2 |81 |44.4 |640|
|AutoSlim-YOLOv5m | 16.7(-31.8%)| 17.8(-16%)| 69(-14.8%)| 42.0(-2.4)| 640|

Expand Down Expand Up @@ -108,7 +116,7 @@ Refer to the paper [Distilling the Knowledge in a Neural Network](https://arxiv.
### 1.4 Pruning

AutoSlim can prune the model automatically, which it can achieve better model accuracy under
limited resource conditions (such as FLOPs, latency, memory footprint, or model size)。In AutoSlim,
limited resource conditions (such as FLOPs, latency, memory footprint, or model size). In AutoSlim,
it can be divided into several steps. The first step is to train a slimmable model for a few epochs
(e.g., 10% to 20% of full training epochs) to quickly get a benchmark performance estimator; Then we
evaluate the trained slimmable model and greedily slim the layer with minimal accuracy drop on a
Expand Down Expand Up @@ -157,14 +165,14 @@ python -m pip install -r requirements.txt

There are two installation methods.

1、Python wheel installation
- Python wheel installation

```sh
cd model_optimizer
python setup.py install
```

2、Developer mode installation
- Developer mode installation

```sh
chmod +x *.sh
Expand Down Expand Up @@ -358,6 +366,32 @@ examples/classifier_imagenet/prototxt/resnet/resnet50_autoslim_search.prototxt
examples/classifier_imagenet/prototxt/resnet/resnet50_autoslim_retrain_100epochs_lr0.4_decay5e05_momentum0.9_ls0.1.prototxt
```

### 3.5 Low-bit quantization

Here is a detailed introduction to the steps to reproduce the quantitative results of our model. The quantization process
can be divided into two steps.

(1) model training

First of all, we need a fully trained model as the base model for quantization. Here you can use the following
configuration file to train a higher-precision model.

```sh
examples/classifier_imagenet/prototxt/resnet/resnet50_distillation.prototxt
```

When training the model, we will use resnet50d as the teacher model to distill the student model.

(2) model quantization

When quantizing, load a trained model, then use the LSQ algorithm to quantize the model. To improve
the performance of the quantized model, we will use the distillation method in the quantization process. The following
configuration file is the configuration of our model quantization.

```sh
examples/classifier_imagenet/prototxt/resnet/resnet50_quantization_lsq_3w4a_first_last_layer_int8_per_tensor_distillation_load_weight_80.754.prototxt
```

## Acknowledgement

We would like to thank them for their excellent open-source work.
Expand Down
Loading