fixed bugs in iewrap.py. change inputs/outputs to dictionary

Author: yas-sim

README.md

@@ -106,7 +106,8 @@ list = getInputs()
  - None
 - *Output*
  - List of input blobs of the loaded model.
- - The list format is: `[ {'name':blobName, 'data':blobData, 'shape':blobShape, 'type':blobType('imgae' or others)}, ... ]`.
+ - The list format is: `{ blobName : { 'data':blobData, 'shape':blobShape, 'type':blobType }, ... }`.
+ - `blobType` is a string. The value can be `image`, or others. If the type is `image`, the `blobData` will go through image preprocess before inferencing.
 
 ~~~python
 list = getOutputs()
@@ -118,7 +119,7 @@ list = getOutputs()
  - None
 - *Output*
  - List of output blobs of the loaded model.
- - The list format is: `[ {'name':blobName, 'shape':blobShape}, ... ]`.
+ - The list format is: `{ blobName : { 'shape':blobShape }, ... }`.
 
 ~~~python
 1. outBlob = blockInfer(ocvImg) # for single input model
@@ -132,14 +133,14 @@ list = getOutputs()
  - You can obtain the input list with `getInputs()` and stuff your input data to infer to `data` element in the dictionary in the list.
  - The dictionary has `type` attribute. If you set `image` to the type, the data stored in `data` is considered as an OpenCV image data and go through image preprocessing before inferencing (resize and transform), otherwise the data in the `data` will be just passed to the Inference Engine without any preprocessing.
  - You must use this style of API when your model has multiple inputs. 
- - e.g. `[{'name': 'data', 'data': 0, 'shape': [1, 3, 224, 224], 'type': 'image'}]`
+ - e.g. `{ 'data' : { 'data': 0, 'shape': [1, 3, 224, 224], 'type': 'image'} }`
 - *Return*
  - `outBlob`: Output result of the inferencing
  - Single output model: `outBlob` contains the data of the output blob.
  - Multiple output model: `outBlob` contains a dictionary which contains the outputs of the model.
  - Key: The name of an output blob
  - Value: The contents of an output blob
- - e.g. `[{'name': 'prob', 'shape': [1, 1000]}]`
+ - e.g. `{ 'prob' : { 'shape': [1, 1000] } }`
 
 ~~~python
 1. infID = asyncInfer(ocvimg) # for single input model
@@ -153,7 +154,7 @@ list = getOutputs()
  - You can obtain the input list with `getInputs()` and stuff your input data to infer to `data` element in the dictionary in the list.
  - The dictionary has `type` attribute. If you set `image` to the type, the data stored in `data` is considered as an OpenCV image data and go through image preprocessing before inferencing (resize and transform), otherwise the data in the `data` will be just passed to the Inference Engine without any preprocessing.
  - You must use this style of API when your model has multiple inputs. 
- - e.g. `[{'name': 'data', 'data': 0, 'shape': [1, 3, 224, 224], 'type': 'image'}]`
+ - e.g. `{ 'data' : { 'data': 0, 'shape': [1, 3, 224, 224], 'type': 'image'} }`
 - *Return*
  - `infID`: ID number of the requested inferencing task
 
@@ -169,7 +170,7 @@ setCallback(callback)
  - Multiple output model: `outBlob` contains a dictionary which contains the outputs of the model.
  - Key: The name of an output blob
  - Value: The contents of an output blob
-  - e.g. `[{'name': 'prob', 'shape': [1, 1000]}]`
+ - e.g. `{ 'prob' : { 'shape': [1, 1000] } }`
 - *Return*
  - None
 

iewrap.py

@@ -51,10 +51,11 @@ def callback(self, status, data):
  outputs = req.outputs
  if self.callbackFunc!=None:
  if(len(outputs)==1):
- self.callbackFunc(id, outputs[self.outputs[0]['name']]) # if the model has only 1 output, return the blob contents in an array
+ firstBlobName = next(iter(self.outputs))
+ self.callbackFunc(id, outputs[firstBlobName]) # if the model has only 1 output, return the blob contents in an array
  else:
- self.callbackFunc(id, outputs)  # if the model has multiple outputs, return the result in dictionary
-  # e.g. ( {'prob': array[], 'data': array[]})
+ self.callbackFunc(id, outputs) # if the model has multiple outputs, return the result in dictionary
+ # e.g. ( {'prob': array[], 'data': array[]})
 
  def imagePreprocess(self, img, shape):
  img = cv2.resize(img, (shape[3], shape[2]))

iewrap_gaze_estimation.py

@@ -14,7 +14,7 @@
 ie_faceLM = iewrap.ieWrapper('./intel/facial-landmarks-35-adas-0002/FP16/facial-landmarks-35-adas-0002.xml', 'CPU')
 ie_gaze = iewrap.ieWrapper('./intel/gaze-estimation-adas-0002/FP16/gaze-estimation-adas-0002.xml', 'CPU')
 
-print('hit ESC to exit.')
+print('Please connect a USB web cam. Hit ESC to exit.')
 while cv2.waitKey(1) != 27: # 27 == ESC
  ret,img = cam.read()
  img = cv2.flip(img, 1) # flip image

prep.bat

@@ -12,6 +12,9 @@ rem Download googlenet-v1 and mobilenet-ssd models with Model Downloader and Mod
 python "%INTEL_OPENVINO_DIR%\deployment_tools\tools\model_downloader\downloader.py" --name googlenet-v1,mobilenet-ssd
 python "%INTEL_OPENVINO_DIR%\deployment_tools\tools\model_downloader\converter.py" --name googlenet-v1,mobilenet-ssd --precisions FP16
 
+rem Download models for gaze estimation demo
+python "%INTEL_OPENVINO_DIR%\deployment_tools\tools\model_downloader\downloader.py" --name face-detection-adas-0001,head-pose-estimation-adas-0001,facial-landmarks-35-adas-0002,gaze-estimation-adas-0002
+
 pip install matplotlib opencv-python numpy
 
 exit /B

prep.sh

@@ -16,4 +16,7 @@ cp $INTEL_OPENVINO_DIR/deployment_tools/inference_engine/samples/python/voc_labe
 python3 $INTEL_OPENVINO_DIR/deployment_tools/tools/model_downloader/downloader.py --name googlenet-v1,mobilenet-ssd
 python3 $INTEL_OPENVINO_DIR/deployment_tools/tools/model_downloader/converter.py --name googlenet-v1,mobilenet-ssd --precisions FP16
 
+# Download models for gaze estimation demo
+python3 $INTEL_OPENVINO_DIR/deployment_tools/tools/model_downloader/downloader.py --name face-detection-adas-0001,head-pose-estimation-adas-0001,facial-landmarks-35-adas-0002,gaze-estimation-adas-0002
+
 pip3 install matplotlib opencv-python numpy