MobileNetV2 is tuned to mobile phone CPUs through a combination of hardware- aware network architecture search (NAS) complemented by the NetAdapt algorithm and then subsequently improved through novel architecture advances.Nov 20, 2019.
[Paper](https://arxiv.org/pdf/1905.02244) Howard, Andrew, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang et al. "Searching for MobileNetV2." In Proceedings of the IEEE International Conference on Computer Vision, pp. 1314-1324. 2019.
@ -38,11 +37,10 @@ The overall network architecture of MobileNetV2 is show below:
- Dataset size: ~125G, 1.2W colorful images in 1000 classes
- Train: 120G, 1.2W images
- Test: 5G, 50000 images
- Train: 120G, 1.2W images
- Test: 5G, 50000 images
- Data format: RGB images.
- Note: Data will be processed in src/dataset.py
- Note: Data will be processed in src/dataset.py
# [Features](#contents)
@ -54,14 +52,13 @@ For FP16 operators, if the input data type is FP32, the backend of MindSpore wil
# [Environment Requirements](#contents)
- Hardware:Ascend
- Prepare hardware environment with Ascend. If you want to try Ascend , please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
- Prepare hardware environment with Ascend. If you want to try Ascend , please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
@ -133,7 +127,7 @@ You can start training using python or shell scripts. The usage of shell scripts
Training result will be stored in the example path. Checkpoints trained by `Ascend` will be stored at `./train/device$i/checkpoint` by default, and training log will be redirected to `./train/device$i/train.log`. Checkpoints trained by `GPU` will be stored in `./train/checkpointckpt_$i` by default, and training log will be redirected to `./train/train.log`.
- Dataset size 224*224 colorful images in 1000 classes
- Train：1,281,167 images
- Test： 50,000 images
- Train：1,281,167 images
- Test： 50,000 images
- Data format：jpeg
- Note：Data will be processed in dataset.py
- Note：Data will be processed in dataset.py
- Download the dataset, the directory structure is as follows:
```
```python
└─dataset
├─ilsvrc # train dataset
└─validation_preprocess # evaluate dataset
```
# [Features](#contents)
## [Mixed Precision](#contents)
@ -60,14 +59,13 @@ For FP16 operators, if the input data type is FP32, the backend of MindSpore wil
# [Environment Requirements](#contents)
- Hardware:Ascend
- Prepare hardware environment with Ascend. If you want to try Ascend , please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
- Prepare hardware environment with Ascend. If you want to try Ascend , please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
Training result will be stored in the example path. Checkpoints will be stored at `./train/device$i/` by default, and training log will be redirected to `./train/device$i/train.log` like followings.
Training result will be stored in the example path. Checkpoints will be stored at `./train/device$i/` by default, and training log will be redirected to `./train/device$i/train.log` like following.
```
```bash
epoch: 1 step: 5004, loss is 4.8995576
epoch: 2 step: 5004, loss is 3.9235563
epoch: 3 step: 5004, loss is 3.833077
@ -153,7 +152,7 @@ You can start training using python or shell scripts. The usage of shell scripts
### Launch
```
```bash
# infer example
shell:
Ascend: sh run_infer.sh Ascend ~/imagenet/val/ ~/train/Resnet50-30_5004.ckpt
@ -163,9 +162,9 @@ You can start training using python or shell scripts. The usage of shell scripts
### Result
Inference result will be stored in the example path, you can find result like the followings in `./eval/infer.log`.
Inference result will be stored in the example path, you can find result like the following in `./eval/infer.log`.
- [Script and Sample Code](#script-and-sample-code)
- [Script Parameters](#script-parameters)
@ -19,7 +19,6 @@
- [Description of Random Situation](#description-of-random-situation)
- [ModelZoo Homepage](#modelzoo-homepage)
# [YOLOv3_ResNet18 Description](#contents)
YOLOv3 network based on ResNet-18, with support for training and evaluation.
@ -30,24 +29,25 @@ YOLOv3 network based on ResNet-18, with support for training and evaluation.
The overall network architecture of YOLOv3 is show below:
And we use ResNet18 as the backbone of YOLOv3_ResNet18. The architecture of ResNet18 has 4 stages. The ResNet architecture performs the initial convolution and max-pooling using 7×7 and 3×3 kernel sizes respectively. Afterward, every stage of the network has different Residual blocks (2, 2, 2, 2) containing two 3×3 conv layers. Finally, the network has an Average Pooling layer followed by a fully connected layer.
And we use ResNet18 as the backbone of YOLOv3_ResNet18. The architecture of ResNet18 has 4 stages. The ResNet architecture performs the initial convolution and max-pooling using 7×7 and 3×3 kernel sizes respectively. Afterward, every stage of the network has different Residual blocks (2, 2, 2, 2) containing two 3×3 conv layers. Finally, the network has an Average Pooling layer followed by a fully connected layer.
# [Dataset](#contents)
Note that you can run the scripts based on the dataset mentioned in original paper or widely used in relevant domain/network architecture. In the following sections, we will introduce how to run the scripts using the related dataset below.
Each row is an image annotation which split by space, the first column is a relative path of image, the others are box and class infomations of the format [xmin,ymin,xmax,ymax,class]. `dataset.py` is the parsing script, we read image from an image path joined by the `image_dir`(dataset directory) and the relative path in `anno_path`(the TXT file path), `image_dir` and `anno_path` are external inputs.
# [Environment Requirements](#contents)
- Hardware（Ascend）
- Prepare hardware environment with Ascend processor. If you want to try Ascend , please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
- Prepare hardware environment with Ascend processor. If you want to try Ascend , please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
After installing MindSpore via the official website, you can start training and evaluation on Ascend as follows:
After installing MindSpore via the official website, you can start training and evaluation on Ascend as follows:
- runing on Ascend
- running on Ascend
```shell script
#run standalone training example
sh run_standalone_train.sh [DEVICE_ID] [EPOCH_SIZE] [MINDRECORD_DIR] [IMAGE_DIR] [ANNO_PATH]
#run distributed training example
sh run_distribute_train.sh [DEVICE_NUM] [EPOCH_SIZE] [MINDRECORD_DIR] [IMAGE_DIR] [ANNO_PATH] [RANK_TABLE_FILE]
#run evaluation example
sh run_eval.sh [DEVICE_ID] [CKPT_PATH] [MINDRECORD_DIR] [IMAGE_DIR] [ANNO_PATH]
```
# [Script Description](#contents)
## [Script and Sample Code](#contents)
```
```python
└── cv
├── README.md // descriptions about all the models
├── mindspore_hub_conf.md // config for mindspore hub
└── yolov3_resnet18
└── yolov3_resnet18
├── README.md // descriptions about yolov3_resnet18
├── scripts
├── scripts
├── run_distribute_train.sh // shell script for distributed on Ascend
├── run_standalone_train.sh // shell script for distributed on Ascend
└── run_eval.sh // shell script for evaluation on Ascend
├── src
├── src
├── dataset.py // creating dataset
├── yolov3.py // yolov3 architecture
├── config.py // parameter configuration
├── config.py // parameter configuration
└── utils.py // util function
├── train.py // training script
├── train.py // training script
└── eval.py // evaluation script
```
## [Script Parameters](#contents)
```
Major parameters in train.py and config.py as follows:
```python
device_num: Use device nums, default is 1.
lr: Learning rate, default is 0.001.
epoch_size: Epoch size, default is 50.
@ -133,24 +131,23 @@ After installing MindSpore via the official website, you can start training and
img_shape: Image height and width used as input to the model.
```
## [Training Process](#contents)
### Training on Ascend
To train the model, run `train.py` with the dataset `image_dir`, `anno_path` and `mindrecord_dir`. If the `mindrecord_dir` is empty, it wil generate [mindrecord](https://www.mindspore.cn/tutorial/training/zh-CN/master/advanced_use/convert_dataset.html) file by `image_dir` and `anno_path`(the absolute image path is joined by the `image_dir` and the relative path in `anno_path`). **Note if `mindrecord_dir` isn't empty, it will use `mindrecord_dir` rather than `image_dir` and `anno_path`.**
- Stand alone mode
```
```bash
sh run_standalone_train.sh 0 50 ./Mindrecord_train ./dataset ./dataset/train.txt
```
The input variables are device id, epoch size, mindrecord directory path, dataset directory path and train TXT file path.
- Distributed mode
```
```bash
sh run_distribute_train.sh 8 150 /data/Mindrecord_train /data /data/train.txt /data/hccl.json
```
@ -158,7 +155,7 @@ To train the model, run `train.py` with the dataset `image_dir`, `anno_path` and
You will get the loss value and time of each step as following:
```
```bash
epoch: 145 step: 156, loss is 12.202981
epoch time: 25599.22742843628, per step time: 164.0976117207454
epoch: 146 step: 156, loss is 16.91706
@ -168,19 +165,20 @@ You will get the loss value and time of each step as following:
epoch: 148 step: 156, loss is 10.431475
epoch time: 23634.241580963135, per step time: 151.50154859591754
epoch: 149 step: 156, loss is 14.665991
epoch time: 24118.8325881958, per step time: 154.60790120638333
epoch time: 24118.8325881958, per step time: 154.60790120638333
epoch: 150 step: 156, loss is 10.779521
epoch time: 25319.57221031189, per step time: 162.30495006610187
```
Note the results is two-classification(person and face) used our own annotations with coco2017, you can change `num_classes` in `config.py` to train your dataset. And we will suport 80 classifications in coco2017 the near future.
Note the results is two-classification(person and face) used our own annotations with coco2017, you can change `num_classes` in `config.py` to train your dataset. And we will support 80 classifications in coco2017 the near future.
## [Evaluation Process](#contents)
### Evaluation on Ascend
To eval, run `eval.py` with the dataset `image_dir`, `anno_path`(eval txt), `mindrecord_dir` and `ckpt_path`. `ckpt_path` is the path of [checkpoint](https://www.mindspore.cn/tutorial/training/en/master/use/save_model.html) file.
```
```bash
sh run_eval.sh 0 yolo.ckpt ./Mindrecord_eval ./dataset ./dataset/eval.txt
```
@ -188,18 +186,18 @@ The input variables are device id, checkpoint path, mindrecord directory path, d
You will get the precision and recall value of each class:
```
```bash
class 0 precision is 88.18%, recall is 66.00%
class 1 precision is 85.34%, recall is 79.13%
```
Note the precision and recall values are results of two-classification(person and face) used our own annotations with coco2017.
zlq2020
4 years ago