In [ ]:

    

# load model
%load_ext autoreload
%autoreload 2

model_filepath='hog_svm_daimler_model.pkl'
from sklearn.externals import joblib
model=joblib.load(model_filepath)


    

In [16]:

    

def visualize_classification(basepaths,dataset,index,hog_image_size,bb):

    images_basepath=basepaths[dataset]

    image_names=os.listdir(images_basepath)
    image_filename=image_names[index]
    image_filepath=os.path.join(images_basepath,image_filename)
    image=io.imread(image_filepath)
    
    row,col,w,h=bb
        
    if dataset=="pedestrian":
        area=image
    else:
        area=skimage.transform.resize(image[row:row+h,col:col+w],hog_image_size,mode="reflect")


    hog,hog_visualization=experiment.my_hog(area,visualise=True)
    prediction=model.predict(hog.reshape(1,-1))

    fig,(ax_image,ax_area,ax_hog)=plt.subplots(1,3,dpi=500)

    ax_image.imshow(image)
    ax_image.set_title("Dataset "+dataset+", File: "+image_filename,fontsize=5)

    if dataset!="pedestrian":
        p = patches.Rectangle((col,row), w, h, fill=False)
        ax_image.add_patch(p)

    ax_area.imshow(area)
    pedestrian_labels={0:"non_pedestrian",1:"pedestrian"}
    ax_area.set_title("Prediction for area: "+pedestrian_labels[prediction[0]],fontsize=5)
    ax_hog.imshow(hog_visualization)
    ax_hog.set_title("Hog visualization for area",fontsize=5)
    plt.tight_layout()
    plt.show()


    

In [17]:

    

# test with a new positive image
import os
import skimage.io as io
import matplotlib.pyplot as plt
import experiment
import skimage
import matplotlib.patches as patches

# set basepaths of each subset of the dataset
basepaths={"pedestrian":"/media/data/datasets/pedestrian/daimler_mono/DaimlerBenchmark/Data/TrainingData/Pedestrians/48x96",
           "non_pedestrian":"/media/data/datasets/pedestrian/daimler_mono/DaimlerBenchmark/Data/TrainingData/NonPedestrians",
           "test":"/media/data/datasets/pedestrian/daimler_mono/DaimlerBenchmark/Data/TestData",}

# set the size of the image patches used to generate the hogs the classifier was trained on.
hog_image_size=(96,48)


## Choose dataset and image index to predict (datasets: "test", "pedestrian", "non_pedestrian"
dataset="test"
index=3006 
# Also pick x,y,w,h position of bounding box to predict if dataset="test" or dataset=="non_pedestrian"
row=160
col=150
w=48
h=96
bb=(row,col,w,h)
# index=0
# row=264
# col=480
visualize_classification(basepaths,dataset,index,hog_image_size,bb)
## END