add code to draw the figures in the paper

draw_figures_in_the_paper/count_pred_all.json

@@ -0,0 +1 @@
+[["on", 0.3155507021783991], ["has", 0.1565315839763612], ["wearing", 0.1142129900996459], ["of", 0.08633462072731068], ["in", 0.06088597329393533], ["near", 0.048708457455777614], ["with", 0.029593467211601], ["behind", 0.02916148095807806], ["holding", 0.026519780634489846], ["above", 0.018631615291349837], ["sitting on", 0.012851992516520664], ["wears", 0.011997655390594263], ["under", 0.010738632257650092], ["riding", 0.01017335656530781], ["in front of", 0.009007475449851856], ["standing on", 0.006532788399001132], ["at", 0.0048658674653327015], ["attached to", 0.004318256638376117], ["carrying", 0.00402116572052577], ["walking on", 0.003762616327153307], ["over", 0.003004633012421612], ["belonging to", 0.0024377514232260863], ["for", 0.00242008655784039], ["looking at", 0.002264314563075614], ["watching", 0.002131025124256269], ["hanging from", 0.0020812423218056703], ["parked on", 0.0018660521434708248], ["laying on", 0.001827510618992942], ["eating", 0.0017070683549995583], ["and", 0.0014983017640776933], ["covering", 0.0013296825944869562], ["using", 0.0013200472133674853], ["between", 0.0012285110927325138], ["covered in", 0.001132157281537807], ["along", 0.0010968275507664143], ["part of", 0.0010020796364249524], ["lying on", 0.0009009081346705102], ["on back of", 0.0008784255787250784], ["to", 0.0008334604668342152], ["mounted on", 0.0007804658706771264], ["walking in", 0.0006841120594824195], ["across", 0.0005845464545812223], ["against", 0.0005106751993319469], ["from", 0.0004914044370930055], ["growing on", 0.00047534546856055435], ["painted on", 0.00042235087240346554], ["made of", 0.00027621425875815997], ["playing", 0.00026657887763868927], ["says", 9.956560490119719e-05], ["flying in", 4.817690559735348e-05]]

draw_figures_in_the_paper/draw_figures.py

@@ -0,0 +1,200 @@
+# -*- coding: UTF-8 -*-
+
+
+import sys
+import os
+import math
+import pickle
+import numpy as np
+import json
+import matplotlib.pyplot as plt
+import matplotlib
+matplotlib.use('Agg')
+
+
+
+
+
+def draw_scatter():
+
+    with open('nocontrained/motifnet_predcls_sg_eval_result_mean_recall.pkl', 'rb') as f:
+        cvpr18 = pickle.load(f)
+    with open('nocontrained/kern_predcls_sg_eval_result_mean_recall.pkl', 'rb') as f:
+        our = pickle.load(f)
+    with open('count_pred_all.json', 'r') as f:
+        data = json.load(f)
+    rel_list = []
+    num_list = []
+    for rel, num in data:
+        rel_list.append(rel)
+        num_list.append(num)
+    num_list = np.array(num_list) * 100
+    cvpr18_list = []
+    our_list = []
+    for r in rel_list:
+        cvpr18_list.append(cvpr18[r]['R@50'])
+        our_list.append(our[r]['R@50'])
+    plt.figure(figsize=(10,5))
+    x = np.arange(50)+1
+    y1 = np.array(cvpr18_list) * 100
+    y2 = np.array(our_list) * 100
+    y3 = y2 - y1
+    plt.scatter(num_list, y3 ,s=30,color='green',marker='o',alpha=0.5)
+
+    plt.xlabel('Relationship proportion (%)', fontsize=14)
+    plt.ylabel('Improvement proportion (%)', fontsize=14)
+
+    plt.grid(True, linestyle='--', axis='y')
+    # plt.legend()
+    plt.tight_layout()
+    plt.savefig('scatter.pdf')
+
+
+
+def draw_difference_compare():
+
+
+
+    with open('nocontrained/motifnet_predcls_sg_eval_result_mean_recall.pkl', 'rb') as f:
+        cvpr18 = pickle.load(f)
+    with open('nocontrained/kern_predcls_sg_eval_result_mean_recall.pkl', 'rb') as f:
+        our = pickle.load(f)
+    with open('count_pred_all.json', 'r') as f:
+        data = json.load(f)
+    rel_list = []
+    num_list = []
+    for rel, num in data:
+        rel_list.append(rel)
+        num_list.append(num)
+    cvpr18_list = []
+    our_list = []
+    for r in rel_list:
+        cvpr18_list.append(cvpr18[r]['R@50'])
+        our_list.append(our[r]['R@50'])
+    plt.figure(figsize=(10,5))
+    x = np.arange(50)+1
+    y1 = np.array(cvpr18_list) * 100
+    y2 = np.array(our_list) * 100
+    y3 = y2 - y1
+    plt.bar(x, y3, alpha=0.9, width = 0.7, facecolor = 'green', edgecolor = 'white', label='SMN', lw=1)
+    y4 = y3.copy()
+    y4[y4>0] = 0
+    plt.bar(x, y4, alpha=0.9, width = 0.7, facecolor = 'red', edgecolor = 'white', label='SMN', lw=1)
+
+    xticks1=rel_list
+    plt.xticks(x,xticks1,fontsize=14,rotation=90) 
+
+    plt.ylabel('R@50 Improvement (%)', fontsize=14) 
+
+    for a,b,c,d in zip(x,y1,y2, y3):
+        plt.text(a, d if d >= 0 else d - 10, '%+-.2f' % d, ha='center', va= 'bottom',fontsize=14, rotation=90) 
+    plt.xlim(0,51)
+    plt.ylim(-5, 35) 
+    plt.ylim(-15, 45)
+    plt.grid(True, linestyle='--', axis='y')
+    # plt.legend()
+    plt.tight_layout()
+    plt.savefig('difference_compare.pdf')
+
+
+
+
+
+
+def draw_compare():
+    with open('nocontrained/motifnet_predcls_sg_eval_result_mean_recall.pkl', 'rb') as f:
+        cvpr18 = pickle.load(f)
+    with open('nocontrained/kern_predcls_sg_eval_result_mean_recall.pkl', 'rb') as f:
+        our = pickle.load(f)
+    with open('count_pred_all.json', 'r') as f:
+        data = json.load(f)
+
+    rel_list = []
+    num_list = []
+    for rel, num in data:
+        rel_list.append(rel)
+        num_list.append(num)
+    cvpr18_list = []
+    our_list = []
+    for r in rel_list:
+        cvpr18_list.append(cvpr18[r]['R@50'])
+        our_list.append(our[r]['R@50'])
+    plt.figure(figsize=(10,5))
+    x = np.arange(50)+1
+    y1 = np.array(cvpr18_list) * 100
+    y2 = np.array(our_list) * 100
+    y3 = y2 - y1
+    plt.bar(x-0.175, y2, alpha=1, width = 0.35, facecolor = 'coral', edgecolor = 'white', label='Ours', lw=1)
+    plt.bar(x+0.175, y1, alpha=1, width = 0.35, facecolor = 'c', edgecolor = 'white', label='SMN', lw=1)
+
+    y4 = y3.copy()
+    y4[y4>0] = 0
+
+    xticks1=rel_list
+
+    plt.xticks(x,xticks1,fontsize=14,rotation=90) 
+
+    plt.ylabel('R@50 (%)', fontsize=14) 
+
+    for a,b,c,d in zip(x,y1,y2, y3):
+        plt.text(a, b+0.5 if b>c else c+0.5, '%+-.2f' % d, ha='center', va= 'bottom',fontsize=14, rotation=90)
+    plt.xlim(0,51)
+    plt.ylim(0, 119) 
+    plt.grid(True, linestyle='--', axis='y')
+    plt.legend(fontsize=14)
+    plt.tight_layout()
+    plt.savefig('compare.pdf')
+
+
+
+
+
+def draw_bar(file_name):
+    with open(file_name, 'r') as f:
+        data = json.load(f)
+    rel_list = []
+    num_list = []
+    for rel, num in data:
+        rel_list.append(rel)
+        num_list.append(num)
+
+
+
+    # in CVPR paper, the font size is 14
+
+    plt.figure(figsize=(10,5))
+
+    x=np.arange(50)+1 
+
+    y=np.array(num_list) * 100
+    xticks1=rel_list 
+
+    plt.bar(x,y,width = 0.7,align='center',color = 'lightcoral',alpha=1, edgecolor = 'white')
+
+    plt.tick_params(labelsize=14)
+    plt.xticks(x,xticks1,fontsize=12,rotation=90) 
+
+    # plt.xlabel('Relationship')
+    plt.ylabel('Proportion (%)', fontsize=14) 
+
+    for a,b in zip(x,y):
+
+        plt.text(a, b+0.5, '%.3f' % b, ha='center', va= 'bottom',fontsize=12, rotation=90)
+
+    plt.ylim(0, 42)
+    plt.xlim(0, 51)
+    plt.grid(True, linestyle='--', axis='y')
+    plt.tight_layout()
+
+    plt.savefig('count_pred_all.pdf')
+
+
+if __name__ == "__main__":
+
+
+    draw_bar('count_pred_all.json')
+
+    draw_compare()
+
+    draw_scatter()
+    draw_difference_compare()