OpenCV  5.0.0-pre
Open Source Computer Vision
Conversion of PyTorch Segmentation Models and Launch with OpenCV

Goals

In this tutorial you will learn how to:

We will explore the above-listed points by the example of the FCN ResNet-50 architecture.

Introduction

The key points involved in the transition pipeline of the PyTorch classification and segmentation models with OpenCV API are equal. The first step is model transferring into ONNX format with PyTorch torch.onnx.export built-in function. Further the obtained .onnx model is passed into cv.dnn.readNetFromONNX, which returns cv.dnn.Net object ready for DNN manipulations.

Practice

In this part we are going to cover the following points:

  1. create a segmentation model conversion pipeline and provide the inference
  2. evaluate and test segmentation models

If you'd like merely to run evaluation or test model pipelines, the "Model Conversion Pipeline" part can be skipped.

Model Conversion Pipeline

The code in this subchapter is located in the dnn_model_runner module and can be executed with the line:

`` python -m dnn_model_runner.dnn_conversion.pytorch.segmentation.py_to_py_fcnresnet50 ``

The following code contains the description of the below-listed steps:

  1. instantiate PyTorch model
  2. convert PyTorch model into .onnx
  3. read the transferred network with OpenCV API
  4. prepare input data
  5. provide inference
  6. get colored masks from predictions
  7. visualize results
# initialize PyTorch FCN ResNet-50 model
original_model = models.segmentation.fcn_resnet50(pretrained=True)
# get the path to the converted into ONNX PyTorch model
full_model_path = get_pytorch_onnx_model(original_model)
# read converted .onnx model with OpenCV API
opencv_net = cv2.dnn.readNetFromONNX(full_model_path)
print("OpenCV model was successfully read. Layer IDs: \n", opencv_net.getLayerNames())
# get preprocessed image
img, input_img = get_processed_imgs("test_data/sem_segm/2007_000033.jpg")
# obtain OpenCV DNN predictions
opencv_prediction = get_opencv_dnn_prediction(opencv_net, input_img)
# obtain original PyTorch ResNet50 predictions
pytorch_prediction = get_pytorch_dnn_prediction(original_model, input_img)
pascal_voc_classes, pascal_voc_colors = read_colors_info("test_data/sem_segm/pascal-classes.txt")
# obtain colored segmentation masks
opencv_colored_mask = get_colored_mask(img.shape, opencv_prediction, pascal_voc_colors)
pytorch_colored_mask = get_colored_mask(img.shape, pytorch_prediction, pascal_voc_colors)
# obtain palette of PASCAL VOC colors
color_legend = get_legend(pascal_voc_classes, pascal_voc_colors)
cv2.imshow('PyTorch Colored Mask', pytorch_colored_mask)
cv2.imshow('OpenCV DNN Colored Mask', opencv_colored_mask)
cv2.imshow('Color Legend', color_legend)
cv2.waitKey(0)

To provide the model inference we will use the below picture from the PASCAL VOC validation dataset:

PASCAL VOC img

The target segmented result is:

PASCAL VOC ground truth

For the PASCAL VOC colors decoding and its mapping with the predicted masks, we also need pascal-classes.txt file, which contains the full list of the PASCAL VOC classes and corresponding colors.

Let's go deeper into each code step by the example of pretrained PyTorch FCN ResNet-50:

# initialize PyTorch FCN ResNet-50 model
original_model = models.segmentation.fcn_resnet50(pretrained=True)
# define the directory for further converted model save
onnx_model_path = "models"
# define the name of further converted model
onnx_model_name = "fcnresnet50.onnx"
# create directory for further converted model
os.makedirs(onnx_model_path, exist_ok=True)
# get full path to the converted model
full_model_path = os.path.join(onnx_model_path, onnx_model_name)
# generate model input to build the graph
generated_input = Variable(
torch.randn(1, 3, 500, 500)
)
# model export into ONNX format
torch.onnx.export(
original_model,
generated_input,
full_model_path,
verbose=True,
input_names=["input"],
output_names=["output"],
opset_version=11
)

The code from this step does not differ from the classification conversion case. Thus, after the successful execution of the above code, we will get models/fcnresnet50.onnx.

# read converted .onnx model with OpenCV API
opencv_net = cv2.dnn.readNetFromONNX(full_model_path)
# read the image
input_img = cv2.imread(img_path, cv2.IMREAD_COLOR)
input_img = input_img.astype(np.float32)
# target image sizes
img_height = input_img.shape[0]
img_width = input_img.shape[1]
# define preprocess parameters
mean = np.array([0.485, 0.456, 0.406]) * 255.0
scale = 1 / 255.0
std = [0.229, 0.224, 0.225]
# prepare input blob to fit the model input:
# 1. subtract mean
# 2. scale to set pixel values from 0 to 1
input_blob = cv2.dnn.blobFromImage(
image=input_img,
scalefactor=scale,
size=(img_width, img_height), # img target size
mean=mean,
swapRB=True, # BGR -> RGB
crop=False # center crop
)
# 3. divide by std
input_blob[0] /= np.asarray(std, dtype=np.float32).reshape(3, 1, 1)

In this step we read the image and prepare model input with cv2.dnn.blobFromImage function, which returns 4-dimensional blob. It should be noted that firstly in cv2.dnn.blobFromImage mean value is subtracted and only then pixel values are scaled. Thus, mean is multiplied by 255.0 to reproduce the original image preprocessing order:

img /= 255.0
img -= [0.485, 0.456, 0.406]
img /= [0.229, 0.224, 0.225]
# set OpenCV DNN input
opencv_net.setInput(preproc_img)
# OpenCV DNN inference
out = opencv_net.forward()
print("OpenCV DNN segmentation prediction: \n")
print("* shape: ", out.shape)
# get IDs of predicted classes
out_predictions = np.argmax(out[0], axis=0)

After the above code execution we will get the following output:

OpenCV DNN segmentation prediction:
* shape: (1, 21, 500, 500)

Each prediction channel out of 21, where 21 represents the number of PASCAL VOC classes, contains probabilities, which indicate how likely the pixel corresponds to the PASCAL VOC class.

original_net.eval()
preproc_img = torch.FloatTensor(preproc_img)
with torch.no_grad():
# obtaining unnormalized probabilities for each class
out = original_net(preproc_img)['out']
print("\nPyTorch segmentation model prediction: \n")
print("* shape: ", out.shape)
# get IDs of predicted classes
out_predictions = out[0].argmax(dim=0)

After the above code launching we will get the following output:

PyTorch segmentation model prediction:
* shape: torch.Size([1, 21, 366, 500])

PyTorch prediction also contains probabilities corresponding to each class prediction.

# convert mask values into PASCAL VOC colors
processed_mask = np.stack([colors[color_id] for color_id in segm_mask.flatten()])
# reshape mask into 3-channel image
processed_mask = processed_mask.reshape(mask_height, mask_width, 3)
processed_mask = cv2.resize(processed_mask, (img_width, img_height), interpolation=cv2.INTER_NEAREST).astype(
np.uint8)
# convert colored mask from BGR to RGB for compatibility with PASCAL VOC colors
processed_mask = cv2.cvtColor(processed_mask, cv2.COLOR_BGR2RGB)

In this step we map the probabilities from segmentation masks with appropriate colors of the predicted classes. Let's have a look at the results:

OpenCV Colored Mask

For the extended evaluation of the models, we can use py_to_py_segm script of the dnn_model_runner module. This module part will be described in the next subchapter.

Evaluation of the Models

The proposed in dnn/samples dnn_model_runner module allows to run the full evaluation pipeline on the PASCAL VOC dataset and test execution for the following PyTorch segmentation models:

This list can be also extended with further appropriate evaluation pipeline configuration.

Evaluation Mode

The below line represents running of the module in the evaluation mode:

python -m dnn_model_runner.dnn_conversion.pytorch.segmentation.py_to_py_segm --model_name <pytorch_segm_model_name>

Chosen from the list segmentation model will be read into OpenCV cv.dnn_Net object. Evaluation results of PyTorch and OpenCV models (pixel accuracy, mean IoU, inference time) will be written into the log file. Inference time values will be also depicted in a chart to generalize the obtained model information.

Necessary evaluation configurations are defined in the test_config.py:

@dataclass
class TestSegmConfig:
frame_size: int = 500
img_root_dir: str = "./VOC2012"
img_dir: str = os.path.join(img_root_dir, "JPEGImages/")
img_segm_gt_dir: str = os.path.join(img_root_dir, "SegmentationClass/")
# reduced val: https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/data/pascal/seg11valid.txt
segm_val_file: str = os.path.join(img_root_dir, "ImageSets/Segmentation/seg11valid.txt")
colour_file_cls: str = os.path.join(img_root_dir, "ImageSets/Segmentation/pascal-classes.txt")

These values can be modified in accordance with chosen model pipeline.

To initiate the evaluation of the PyTorch FCN ResNet-50, run the following line:

python -m dnn_model_runner.dnn_conversion.pytorch.segmentation.py_to_py_segm --model_name fcnresnet50

Test Mode

The below line represents running of the module in the test mode, which provides the steps for the model inference:

python -m dnn_model_runner.dnn_conversion.pytorch.segmentation.py_to_py_segm --model_name <pytorch_segm_model_name> --test True --default_img_preprocess <True/False> --evaluate False

Here default_img_preprocess key defines whether you'd like to parametrize the model test process with some particular values or use the default values, for example, scale, mean or std.

Test configuration is represented in test_config.py TestSegmModuleConfig class:

@dataclass
class TestSegmModuleConfig:
segm_test_data_dir: str = "test_data/sem_segm"
test_module_name: str = "segmentation"
test_module_path: str = "segmentation.py"
input_img: str = os.path.join(segm_test_data_dir, "2007_000033.jpg")
model: str = ""
frame_height: str = str(TestSegmConfig.frame_size)
frame_width: str = str(TestSegmConfig.frame_size)
scale: float = 1.0
mean: List[float] = field(default_factory=lambda: [0.0, 0.0, 0.0])
std: List[float] = field(default_factory=list)
crop: bool = False
rgb: bool = True
classes: str = os.path.join(segm_test_data_dir, "pascal-classes.txt")

The default image preprocessing options are defined in default_preprocess_config.py:

pytorch_segm_input_blob = {
"mean": ["123.675", "116.28", "103.53"],
"scale": str(1 / 255.0),
"std": ["0.229", "0.224", "0.225"],
"crop": "False",
"rgb": "True"
}

The basis of the model testing is represented in samples/dnn/segmentation.py. segmentation.py can be executed autonomously with provided converted model in --input and populated parameters for cv2.dnn.blobFromImage.

To reproduce from scratch the described in "Model Conversion Pipeline" OpenCV steps with dnn_model_runner execute the below line:

python -m dnn_model_runner.dnn_conversion.pytorch.segmentation.py_to_py_segm --model_name fcnresnet50 --test True --default_img_preprocess True --evaluate False