In [2]:
%matplotlib inline
import sys
sys.path.append("../src")
from graphprobe import loadviz, summarize
from collections import OrderedDict
import numpy as np
import matplotlib.pyplot as plt
In [3]:
category_colors = OrderedDict([
('scene', '#3288bd'),
('object', '#99d594'),
('part', '#e6f598'),
('material', '#fee08b'),
('texture', '#fc8d59'),
('color', '#d53e4f'),
('total', '#aaaaaa')])
threshold = 0.05
In [4]:
directory = "../dissection_test/vgg_material_full_conv/"
blobs = ["fc6-conv", "fc7-conv", "fc8-conv"]
In [5]:
data = []
categories = set(category_colors.keys())
# infer variables$
# use blob name as labels if label if not specified$
# save file as directories/graph.png$
for blob in blobs:
stats = summarize(loadviz(directory, blob), threshold,
top_only=True)
data.append(stats)
categories.update(stats.keys())
x = range(1, len(data) + 1)
maxval = 0
plt.figure(num=None, figsize=(12, 8))
for cat in category_colors.keys():
if cat not in categories:
continue
if cat == "total":
continue
dat = [d[cat] for d in data]
maxval = max(maxval, max(dat))
plt.plot(x, dat, 'o-' if cat != 'total' else 's--',
color=category_colors[cat], label=cat)
plt.xticks(x, blobs)
plt.margins(0.1)
plt.legend(loc="upper left")
plt.ylim(-maxval * 0.05, maxval * 1.5)
ax = plt.gca()
ax.yaxis.grid(True)
for side in ['top', 'bottom', 'right', 'left']:
ax.spines[side].set_visible(False)
ax.xaxis.set_ticks_position('bottom')
plt.title("vgg_material number of detectors")
plt.ylabel('portion of units alinged to a category concept')
file_path = '/home/nakamura/network_dissection/NetDissect/research/vgg_material_num_detectors_1027.ipynb'
plt.text(2.0,0.8,file_path)
plt.savefig(directory + "graph.pdf")
plt.show()
In [10]:
[d["object"] for d in data]
Out[10]:
In [ ]: