Fix resample, add report and result_gen file

Author: hardik01shah

.gitignore

@@ -12,4 +12,5 @@ Assignment_4/seg-net/lib/core/__pycache__/*
 Assignment_4/seg-net/lib/utils/__pycache__/*
 Assignment_4/seg-net/lib/dataset/__pycache__/*
 Assignment_5/code/__pycache__/*
+Assignment_5/results/*
 Submissions/*

Assignment_5/Assignment_5 Report - Hardik.pdf

@@ -223,9 +223,9 @@ def condensation_tracker(video_name, params):
         "hist_bin": 16,
         "alpha": 0,
         "sigma_observe": 0.1,
-        "model": 0,
-        "num_particles": 30,
-        "sigma_position": 15,
+        "model": 1,
+        "num_particles": 300,
+        "sigma_position": 7,
         "sigma_velocity": 1,
         "initial_velocity": (1, 10)
     }

Assignment_5/code/condensation_tracker.py

@@ -0,0 +1,239 @@
+import cv2
+import os
+import numpy as np
+import matplotlib.pyplot as plt
+from matplotlib.widgets import RectangleSelector
+from matplotlib import patches
+
+from color_histogram import color_histogram
+from propagate import propagate
+from observe import observe
+from resample import resample
+from estimate import estimate
+
+
+top_left = []
+bottom_right = []
+
+def line_select_callback(clk, rls):
+    print(clk.xdata, clk.ydata)
+    global top_left
+    global bottom_right
+    top_left = (int(clk.xdata), int(clk.ydata))
+    bottom_right = (int(rls.xdata), int(rls.ydata))
+
+
+def onkeypress(event):
+    global top_left
+    global bottom_right
+    global img
+    if event.key == 'q':
+        print('final bbox', top_left, bottom_right)
+        plt.close()
+
+
+def toggle_selector(event):
+    toggle_selector.RS.set_active(True)
+
+
+def condensation_tracker(video_name, params):
+    '''
+    video_name - video name
+    params - parameters
+        - draw_plats {0, 1} draw output plots throughout
+        - hist_bin   1-255 number of histogram bins for each color: proper values 4,8,16
+        - alpha      number in [0,1]; color histogram update parameter (0 = no update)
+        - sigma_position   std. dev. of system model position noise
+        - sigma_observe    std. dev. of observation model noise
+        - num_particles    number of particles
+        - model            {0,1} system model (0 = no motion, 1 = constant velocity)
+    if using model = 1 then the following parameters are used:
+        - sigma_velocity   std. dev. of system model velocity noise
+        - initial_velocity initial velocity to set particles to
+    '''
+    # Choose video
+    if video_name == "video1.avi":
+        first_frame = 10
+        last_frame = 42
+    elif video_name == "video2.avi":
+        first_frame = 3
+        last_frame = 40
+    elif video_name == "video3.avi":
+        first_frame = 1
+        last_frame = 60
+
+    # Change this to where your data is
+    data_dir = './data/'
+    video_path = os.path.join(data_dir, video_name)
+
+    vidcap = cv2.VideoCapture(video_path)
+    vidcap.set(1, first_frame)
+    ret, first_image = vidcap.read()
+
+    fig, ax = plt.subplots(1)
+    image = first_image
+    frame_height = first_image.shape[0]
+    frame_width = first_image.shape[1]
+
+    first_image = cv2.cvtColor(first_image, cv2.COLOR_BGR2RGB)
+    ax.imshow(first_image)
+
+    toggle_selector.RS = RectangleSelector(
+            ax, line_select_callback,
+            useblit=True,
+            button=[1], minspanx=5, minspany=5,
+            spancoords='pixels', interactive=True
+        )
+    bbox = plt.connect('key_press_event', toggle_selector)
+    key = plt.connect('key_press_event', onkeypress)
+    plt.title("Draw a box then press 'q' to continue")
+    plt.show()
+
+    bbox_width = bottom_right[0] - top_left[0]
+    bbox_height = bottom_right[1] - top_left[1]
+
+    # Get initial color histogram
+    # === implement fuction color_histogram() ===
+    hist = color_histogram(top_left[0], top_left[1], bottom_right[0], bottom_right[1],
+                first_image, params["hist_bin"])
+    # ===========================================
+
+    state_length = 2
+    if(params["model"] == 1):
+        state_length = 4
+
+    # a priori mean state
+    mean_state_a_priori = np.zeros([last_frame - first_frame + 1, state_length])
+    mean_state_a_posteriori = np.zeros([last_frame - first_frame + 1, state_length])
+    # bounding box centre
+    mean_state_a_priori[0, 0:2] = [(top_left[0] + bottom_right[0])/2., (top_left[1] + bottom_right[1])/2.]
+
+    if params["model"] == 1:
+        # use initial velocity
+        mean_state_a_priori[0, 2:4] = params["initial_velocity"]
+
+    # Initialize Particles
+    particles = np.tile(mean_state_a_priori[0], (params["num_particles"], 1))
+    particles_w = np.ones([params["num_particles"], 1]) * 1./params["num_particles"]
+
+    fig, ax = plt.subplots(1)
+    im = ax.imshow(first_image)
+    plt.ion()
+
+    for i in range(last_frame - first_frame + 1):
+        t = i + first_frame
+
+        # Propagate particles
+        # === Implement function propagate() ===
+        particles = propagate(particles, frame_height, frame_width, params)
+        # ======================================
+
+        # Estimate
+        # === Implement function estimate() ===
+        mean_state_a_priori[i, :] = estimate(particles, particles_w)
+        # ======================================
+
+        # Get frame
+        ret, frame = vidcap.read()
+        frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+
+        # Draw
+        if params["draw_plots"]:
+            ax.set_title("Frame: %d" % t)
+            im.set_data(frame)
+            to_remove = []
+
+            # Plot a priori particles
+            new_plot = ax.scatter(particles[:, 0], particles[:, 1], color='blue', s=2)
+            to_remove.append(new_plot)
+
+            # Plot a priori estimation
+            for j in range(i-1, -1, -1):
+                lwidth = 30 - 3 * (i-j)
+                if lwidth > 0:
+                    new_plot = ax.scatter(mean_state_a_priori[j+1, 0], mean_state_a_priori[j+1, 1], color='blue', s=lwidth)
+                    to_remove.append(new_plot)
+                if j != i:
+                    new_plot = ax.plot([mean_state_a_priori[j, 0], mean_state_a_priori[j+1, 0]], 
+                                       [mean_state_a_priori[j, 1], mean_state_a_priori[j+1, 1]], color='blue')
+                    to_remove.append(new_plot[0])
+
+            # Plot a priori bounding box
+            if not np.any(np.isnan(mean_state_a_priori[i, :])):
+                patch = ax.add_patch(patches.Rectangle((mean_state_a_priori[i, 0] - 0.5 * bbox_width, mean_state_a_priori[i, 1] - 0.5 * bbox_height),
+                                                        bbox_width, bbox_height, fill=False, edgecolor='blue', lw=2))
+                to_remove.append(patch)
+
+        # Observe
+        # === Implement function observe() ===
+        particles_w = observe(particles, frame, bbox_height, bbox_width, params["hist_bin"], hist, params["sigma_observe"])
+        # ====================================
+
+        # Update estimation
+        mean_state_a_posteriori[i, :] = estimate(particles, particles_w)
+
+        # Update histogram color model                   
+        hist_crrent = color_histogram(min(max(0, round(mean_state_a_posteriori[i, 0]-0.5*bbox_width)), frame_width-1),
+                                      min(max(0, round(mean_state_a_posteriori[i, 1]-0.5*bbox_height)), frame_height-1),
+                                      min(max(0, round(mean_state_a_posteriori[i, 0]+0.5*bbox_width)), frame_width-1),
+                                      min(max(0, round(mean_state_a_posteriori[i, 1]+0.5*bbox_height)), frame_height-1),
+                                      frame, params["hist_bin"])
+
+        hist = (1 - params["alpha"]) * hist + params["alpha"] * hist_crrent
+
+        if params["draw_plots"]:
+            # Plot updated estimation
+            for j in range(i-1, -1, -1):
+                lwidth = 30 - 3 * (i-j)
+                if lwidth > 0:
+                    new_plot = ax.scatter(mean_state_a_posteriori[j+1, 0], mean_state_a_posteriori[j+1, 1], color='red', s=lwidth)
+                    to_remove.append(new_plot)
+                if j != i:
+                    new_plot = ax.plot([mean_state_a_posteriori[j, 0], mean_state_a_posteriori[j+1, 0]], 
+                                       [mean_state_a_posteriori[j, 1], mean_state_a_posteriori[j+1, 1]], color='red')
+                    to_remove.append(new_plot[0])
+            
+            # Plot updated bounding box
+            if not np.any(np.isnan(mean_state_a_posteriori[i, :])):
+                patch = ax.add_patch(patches.Rectangle((mean_state_a_posteriori[i, 0] - 0.5 * bbox_width, mean_state_a_posteriori[i, 1] - 0.5 * bbox_height),
+                                                        bbox_width, bbox_height, fill=False, edgecolor='red', lw=2))
+                to_remove.append(patch)
+
+
+        # RESAMPLE PARTICLES
+        # === Implement function resample() ===
+        particles, particles_w = resample(particles, particles_w)
+        # =====================================
+
+        if params["draw_plots"] and t != last_frame:
+            
+            plt.pause(0.2)
+            if t == 28:
+                plt.savefig(params["save_dir"] + "_frame_%d.png" % t) # Save frame
+            # plt.savefig(params["save_dir"] + "frame_%d.png" % t) # Save frame
+            # Remove previous element from plot
+            for e in to_remove:
+                e.remove()
+
+    plt.ioff()
+    plt.show()
+
+
+if __name__ == "__main__":
+    video_name = 'video3.avi'
+    params = {
+        "save_dir": "./results/vid3_",
+        "draw_plots": 1,
+        "hist_bin": 16,
+        "alpha": 0.8,
+        "sigma_observe": 0.1,
+        "model": 1,
+        "num_particles": 300,
+        "sigma_position": 7,
+        "sigma_velocity": 1,
+        "initial_velocity": (1, 10)
+    }
+    params["save_dir"] = params["save_dir"] + f"alpha{params['alpha']}_histbin{params['hist_bin']}_numparticles{params['num_particles']}"
+    # if not os.path.exists(params["save_dir"]):
+    #     os.makedirs(params["save_dir"])
+    condensation_tracker(video_name, params)

Assignment_5/code/condensation_tracker_results.py

@@ -3,12 +3,10 @@
 def resample(particles, particles_w):
     ind = np.random.choice(len(particles), len(particles), replace=True, p=particles_w[:,0])
     resampled_particles = particles[ind,:]
+    resampled_particles_w = particles_w[ind,:]
 
-    # chosen indices in old particle array
-    indx_old = np.unique(ind)
-
-    resampled_particles_w = np.zeros([particles.shape[0], 1]) / particles.shape[0]
-    resampled_particles_w[indx_old,:] = particles_w[indx_old,:]
+    # normalize weights
+    resampled_particles_w = resampled_particles_w/np.sum(resampled_particles_w)
 
     assert resampled_particles.shape == particles.shape
     assert resampled_particles_w.shape == particles_w.shape