# EASE : Use POT to Quantize yolo-v3-tf Public Model
###### tags: `POT` `EASE`
## Use OpenVINO dockerhub image
```
docker run -it -v ~/Downloads:/mnt -u root --rm openvino/ubuntu20_dev:latest
```
## Run Accuracy Checker and POT
In ubuntu20_dev docker image,
#### 0. Download COCO 2017 trainval dataset and annotation
```=bash
cd /home/openvino
apt update
apt install unzip
mkdir coco_dataset
cd coco_dataset
curl http://images.cocodataset.org/zips/val2017.zip -o val2017.zip
unzip val2017.zip
curl http://images.cocodataset.org/annotations/annotations_trainval2017.zip -o trainval_2017.zip
unzip trainval_2017.zip
```
##### coco_dataset content
```
coco_dataset/
|--annotations/
|-- captions_train2017.json
|-- captions_val2017.json
|-- instances_train2017.json
|-- instances_val2017.json
|-- person_keypoints_train2017.json
`-- person_keypoints_val2017.json
|-- val2017/
|-- 000000042102.jpg
|-- 000000060102.jpg
|-- 000000245102.jpg
...
`-- 000000364102.jpg
```
#### 1. Download yolo-v3-tf
```=bash
omz_downloader.py --name yolo-v3-tf -o /home/openvino/openvino_models
```
#### 2. Convert yolo-v3-tf to IR
```=bash
omz_converter.py --name yolo-v3-tf -d /home/openvino/openvino_models -o /home/openvino/openvino_models
```
#### 3. Run Accuracy Checker on yolo-v3-tf
```=bash
accuracy_check -c yolo-v3-tf-int8.yml -m /home/openvino/openvino_models/public/yolo-v3-tf/FP32
```
#### 4. Run POT on yolo-v3-tf
```=bash
pot -c yolo-v3-tf-int8.json -e
```
#### 5. Copy yolo-v3-tf FP16-INT8 IR
```=bash
mkdir /home/openvino/openvino_models/public/yolo-v3-tf/FP16-INT8/
cp -ar results/yolo-v3-tf_DefaultQuantization/2*/optimized/* /home/openvino/openvino_models/public/yolo-v3-tf/FP16-INT8/
#### Note : Copy openvino_models foler to /mnt folder. They will be accessable in ~/Downloads folder in the host and /mnt in the container.
```
## Reference
### yolo-v3-tf-int8.yml
```
models:
- name: yolo_v3_tf
launchers:
- framework: dlsdk
device: CPU
adapter:
type: yolo_v3
anchors: "10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326"
num: 9
coords: 4
classes: 80
anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]]
outputs:
- conv2d_58/Conv2D/YoloRegion
- conv2d_66/Conv2D/YoloRegion
- conv2d_74/Conv2D/YoloRegion
datasets:
- name: ms_coco_detection_80_class_without_background
annotation_conversion:
converter: mscoco_detection
annotation_file: /home/openvino/coco_dataset/annotations/instances_val2017.json
data_source: /home/openvino/coco_dataset/val2017
preprocessing:
- type: resize
size: 416
postprocessing:
- type: resize_prediction_boxes
- type: filter
apply_to: prediction
min_confidence: 0.001
remove_filtered: True
- type: nms
overlap: 0.5
- type: clip_boxes
apply_to: prediction
metrics:
- type: map
integral: 11point
ignore_difficult: true
presenter: print_scalar
reference: 0.6227
- type: coco_precision
max_detections: 100
threshold: 0.5
reference: 0.677
```
### yolo-v3-tf-int8.json
```
{
"model": {
"model_name": "yolo-v3-tf",
"model": "/home/openvino/openvino_models/public/yolo-v3-tf/FP16/yolo-v3-tf.xml",
"weights": "/home/openvino/openvino_models/public/yolo-v3-tf/FP16/yolo-v3-tf.bin"
},
"engine": {
"config": "/home/openvino/yolo-v3-tf-int8.yml"
},
"compression": {
"algorithms": [
{
"name": "DefaultQuantization",
"params": {
"preset": "performance",
"stat_subset_size": 300
}
}
]
}
}
```
### accuracy_checker log
```
accuracy_check -c /mnt/yolo-v3-tf-int8.yml -m /mnt/openvino_models/public/yolo-v3-tf/FP32/yolo-v3-tf.xml
Processing info:
model: yolo_v3_tf
launcher: openvino
device: CPU
dataset: ms_coco_detection_80_class_without_background
OpenCV version: 4.5.5
Annotation conversion for ms_coco_detection_80_class_without_background dataset has been started
Parameters to be used for conversion:
converter: mscoco_detection
annotation_file: /mnt/coco_dataset/annotations/instances_val2017.json
Total annotations size: 5000
100 / 5000 processed in 0.602s
200 / 5000 processed in 0.675s
300 / 5000 processed in 0.565s
400 / 5000 processed in 0.573s
500 / 5000 processed in 0.544s
600 / 5000 processed in 0.539s
700 / 5000 processed in 0.529s
800 / 5000 processed in 0.594s
900 / 5000 processed in 0.516s
1000 / 5000 processed in 0.580s
1100 / 5000 processed in 0.555s
1200 / 5000 processed in 0.508s
1300 / 5000 processed in 0.521s
1400 / 5000 processed in 0.564s
1500 / 5000 processed in 0.534s
1600 / 5000 processed in 0.520s
1700 / 5000 processed in 0.519s
1800 / 5000 processed in 0.530s
1900 / 5000 processed in 0.583s
2000 / 5000 processed in 0.568s
2100 / 5000 processed in 0.546s
2200 / 5000 processed in 0.531s
2300 / 5000 processed in 0.577s
2400 / 5000 processed in 0.561s
2500 / 5000 processed in 0.581s
2600 / 5000 processed in 0.568s
2700 / 5000 processed in 0.571s
2800 / 5000 processed in 0.563s
2900 / 5000 processed in 0.568s
3000 / 5000 processed in 0.563s
3100 / 5000 processed in 0.566s
3200 / 5000 processed in 0.561s
3300 / 5000 processed in 0.567s
3400 / 5000 processed in 0.560s
3500 / 5000 processed in 0.567s
3600 / 5000 processed in 0.560s
3700 / 5000 processed in 0.562s
3800 / 5000 processed in 0.553s
3900 / 5000 processed in 0.605s
4000 / 5000 processed in 0.549s
4100 / 5000 processed in 0.529s
4200 / 5000 processed in 0.527s
4300 / 5000 processed in 0.514s
4400 / 5000 processed in 0.531s
4500 / 5000 processed in 0.523s
4600 / 5000 processed in 0.541s
4700 / 5000 processed in 0.495s
4800 / 5000 processed in 0.494s
4900 / 5000 processed in 0.501s
5000 / 5000 processed in 0.492s
5000 objects processed in 27.544 seconds
Annotation conversion for ms_coco_detection_80_class_without_background dataset has been finished
IE version: 2022.1.0-7019-cdb9bec7210-releases/2022/1
Loaded CPU plugin version:
CPU - openvino_intel_cpu_plugin: 2022.1.2022.1.0-7019-cdb9bec7210-releases/2022/1
Found model /mnt/openvino_models/public/yolo-v3-tf/FP32/yolo-v3-tf.xml
Found weights /mnt/openvino_models/public/yolo-v3-tf/FP32/yolo-v3-tf.bin
Input info:
Node name: input_1
Tensor names: input_1:0, input_1
precision: f32
shape: (1, 416, 416, 3)
Output info
Node name: conv2d_58/Conv2D/YoloRegion/sink_port_0
Tensor names: conv2d_58/Conv2D/YoloRegion:0, conv2d_58/Conv2D/YoloRegion
precision: f32
shape: (1, 255, 13, 13)
Node name: conv2d_66/Conv2D/YoloRegion/sink_port_0
Tensor names: conv2d_66/Conv2D/YoloRegion, conv2d_66/Conv2D/YoloRegion:0
precision: f32
shape: (1, 255, 26, 26)
Node name: conv2d_74/Conv2D/YoloRegion/sink_port_0
Tensor names: conv2d_74/Conv2D/YoloRegion:0, conv2d_74/Conv2D/YoloRegion
precision: f32
shape: (1, 255, 52, 52)
5000 objects processed in 1673.242 seconds
map: 62.27%
coco_precision: 67.70%
```
### pot log
```
pot -c /mnt/yolo-v3-tf-int8.json -e
INFO:openvino.tools.pot.app.run:Output log dir: ./results/yolo-v3-tf_DefaultQuantization/2022-04-30_08-39-57
INFO:openvino.tools.pot.app.run:Creating pipeline:
Algorithm: DefaultQuantization
Parameters:
preset : performance
stat_subset_size : 300
target_device : ANY
model_type : None
dump_intermediate_model : False
inplace_statistics : True
exec_log_dir : ./results/yolo-v3-tf_DefaultQuantization/2022-04-30_08-39-57
===========================================================================
IE version: 2022.1.0-7019-cdb9bec7210-releases/2022/1
Loaded CPU plugin version:
CPU - openvino_intel_cpu_plugin: 2022.1.2022.1.0-7019-cdb9bec7210-releases/2022/1
Annotation conversion for ms_coco_detection_80_class_without_background dataset has been started
Parameters to be used for conversion:
converter: mscoco_detection
annotation_file: /mnt/coco_dataset/annotations/instances_val2017.json
Total annotations size: 5000
100 / 5000 processed in 0.566s
200 / 5000 processed in 0.564s
300 / 5000 processed in 0.640s
400 / 5000 processed in 0.591s
500 / 5000 processed in 0.603s
600 / 5000 processed in 0.563s
700 / 5000 processed in 0.565s
800 / 5000 processed in 0.567s
900 / 5000 processed in 0.569s
1000 / 5000 processed in 0.571s
1100 / 5000 processed in 0.574s
1200 / 5000 processed in 0.570s
1300 / 5000 processed in 0.610s
1400 / 5000 processed in 0.588s
1500 / 5000 processed in 0.581s
1600 / 5000 processed in 0.577s
1700 / 5000 processed in 0.576s
1800 / 5000 processed in 0.587s
1900 / 5000 processed in 0.586s
2000 / 5000 processed in 0.579s
2100 / 5000 processed in 0.592s
2200 / 5000 processed in 0.626s
2300 / 5000 processed in 0.585s
2400 / 5000 processed in 0.584s
2500 / 5000 processed in 0.581s
2600 / 5000 processed in 0.580s
2700 / 5000 processed in 0.581s
2800 / 5000 processed in 0.579s
2900 / 5000 processed in 0.586s
3000 / 5000 processed in 0.622s
3100 / 5000 processed in 0.589s
3200 / 5000 processed in 0.576s
3300 / 5000 processed in 0.585s
3400 / 5000 processed in 0.574s
3500 / 5000 processed in 0.572s
3600 / 5000 processed in 0.570s
3700 / 5000 processed in 0.570s
3800 / 5000 processed in 0.578s
3900 / 5000 processed in 0.608s
4000 / 5000 processed in 0.568s
4100 / 5000 processed in 0.566s
4200 / 5000 processed in 0.564s
4300 / 5000 processed in 0.561s
4400 / 5000 processed in 0.560s
4500 / 5000 processed in 0.560s
4600 / 5000 processed in 0.558s
4700 / 5000 processed in 0.565s
4800 / 5000 processed in 0.594s
4900 / 5000 processed in 0.565s
5000 / 5000 processed in 0.554s
5000 objects processed in 28.983 seconds
Annotation conversion for ms_coco_detection_80_class_without_background dataset has been finished
INFO:openvino.tools.pot.pipeline.pipeline:Inference Engine version: 2022.1.0-7019-cdb9bec7210-releases/2022/1
INFO:openvino.tools.pot.pipeline.pipeline:Model Optimizer version: 2022.1.0-7019-cdb9bec7210-releases/2022/1
INFO:openvino.tools.pot.pipeline.pipeline:Post-Training Optimization Tool version: 2022.1.0-7019-cdb9bec7210-releases/2022/1
INFO:openvino.tools.pot.statistics.collector:Start computing statistics for algorithms : DefaultQuantization
INFO:openvino.tools.pot.statistics.collector:Computing statistics finished
INFO:openvino.tools.pot.pipeline.pipeline:Start algorithm: DefaultQuantization
INFO:openvino.tools.pot.algorithms.quantization.default.algorithm:Start computing statistics for algorithm : ActivationChannelAlignment
INFO:openvino.tools.pot.algorithms.quantization.default.algorithm:Computing statistics finished
INFO:openvino.tools.pot.algorithms.quantization.default.algorithm:Start computing statistics for algorithms : MinMaxQuantization,FastBiasCorrection
INFO:openvino.tools.pot.algorithms.quantization.default.algorithm:Computing statistics finished
INFO:openvino.tools.pot.pipeline.pipeline:Finished: DefaultQuantization
===========================================================================
INFO:openvino.tools.pot.pipeline.pipeline:Evaluation of generated model
INFO:openvino.tools.pot.engines.ac_engine:Start inference on the whole dataset
Total dataset size: 5000
1000 / 5000 processed in 120.165s
2000 / 5000 processed in 124.478s
3000 / 5000 processed in 122.800s
4000 / 5000 processed in 123.251s
5000 / 5000 processed in 122.487s
5000 objects processed in 613.181 seconds
INFO:openvino.tools.pot.engines.ac_engine:Inference finished
INFO:openvino.tools.pot.app.run:map : 0.6196563903778992
INFO:openvino.tools.pot.app.run:coco_precision : 0.6766990257494002
```
Sign in with Wallet
Connect another wallet