\( \newcommand{\matr}[1] {\mathbf{#1}} \newcommand{\vertbar} {\rule[-1ex]{0.5pt}{2.5ex}} \newcommand{\horzbar} {\rule[.5ex]{2.5ex}{0.5pt}} \)
deepdream of a sidewalk

Experiment 2

Searching for a image classification fail case by varying illumination.

Using the Multiple Light Source Dataset (mls dataset) to see what effect varying illumination has on classification accuracy.

In the end, it turns out that many of the images in the mls dataset are not applicable to a model trained on ImageNet, as the ImageNet classes don’t cover the types of objects that appear. Experiment 2.0.1 will try to make specific crops of objects that I know contain objects that are ImageNet classes.

Dataset

The dataset consists of the following 24 scenes, repeated in 18 different illuminations.

scenes_overview.png

Method

Take each of the 24x18 preview images, carry out 5 crops (4 corners + center) then pass to ResNet for classification. Record the classification. Compare classifications for for the same scene-crop pair for the 18 different illuminations. I am interested to see if the network classifications are invariant to illumination. The degree of illumination dependency for a scene-crop is measured by tallying the 18 classifications and calculating the entropy for this histogram/distribution. A low entropy means there is low diversity in classifications, and low diversity is what I intent to represent the idea that classification has low dependency on illumination. There are multiple ways one can measure diversity; entropy seems fine for this situtaion. (Diversity measures is an interesting topic in itself; Tom Leinster’s ideas on diversity are great, for example: https://arxiv.org/abs/2012.02113. I try to find any excuse at all to read something by Tom Leinster.)

The code below carries out this idea. The narrative picks up up again afterwards to discuss the (not very useful) results.

import tempfile
import zipfile
import urllib
import numpy as np
import torch
import pathlib
import torchvision as tv
import torchvision.datasets
import torchvision.models
import torchvision.transforms
import pandas as pd
from icecream import ic
import json
import xarray as xr
import matplotlib.pyplot as plt

with open('./resources/imagenet-simple-labels.json') as f:
    labels = json.load(f)
    
def class_id_to_label(cid):
    assert int(cid) == cid
    cid = int(cid)
    return labels[cid]

import IPython
def imshow(img):
    """Show image. 
    
    Image is a HWC numpy array with values in the range 0-1."""
    img = img*255
    img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
    # cv2 imencode takes images in HWC dimension order.
    _,ret = cv2.imencode('.jpg', img) 
    i = IPython.display.Image(data=ret)
    IPython.display.display(i) 
    
    
def imlist(images, labels=None, use_tabs=False):
    if md_export_mode:
        print("Skipping ipyplot image print for markdown export. The output"
              " produces HTML that either Jupyter Lab fails to export correctly,"
              " or Hugo fails to render correctly. Skipping for now.")
        return
    if use_tabs:
        ipyplot.plot_class_tabs(images, labels, max_imgs_per_tab=300)
    else:
        ipyplot.plot_images(images, labels)

# Choose CPU or GPU.
device = torch.device('cuda:0')
#device = "cpu"

# Choose small or large (standard) model variant
#model_name = "resnet18"
model_name = 'resnet50'
def model_fctn():
    if model_name == 'resnet18':
        return tv.models.resnet18(pretrained=True)
    elif model_name == 'resnet50':
        return tv.models.resnet50(pretrained=True)
model = model_fctn()
state = torch.hub.load_state_dict_from_url(tv.models.resnet.model_urls[model_name])
model.load_state_dict(state)
model = model.to(device)
model.eval()

IMG_SHAPE = (224, 224, 3)
BATCH_SIZE = 4
NUM_FC_CHANNELS = 512 if model_name == 'resnet18' else 2048

ds_path = pathlib.Path('resources/exp_2/mls_dataset')
def is_empty(path):
    return not any(path.iterdir())
is_downloaded = ds_path.is_dir() and not is_empty(ds_path)
if not is_downloaded:
    ds_path.mkdir(exist_ok=True)
    zip_path, _ = urllib.request.urlretrieve('ftp://vis.iitp.ru/mls-dataset/images_preview.zip')
    with zipfile.ZipFile(zip_path, "r") as f:
        f.extractall(ds_path)

# Transforms
normalize_transform =  tv.transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
crop = tv.transforms.FiveCrop(size=IMG_SHAPE[0])
pre_norm_transform = tv.transforms.Compose([crop])
norm_transform = tv.transforms.Compose([normalize_transform])

# Data details
num_scenes = 24
ignore_scenes = {1,} # 0: the Macbeth chart
scenes = [s for s in range(1, num_scenes+1) if s not in ignore_scenes]
crops = ['topleft', 'topright', 'bottomleft', 'bottomright', 'center']
illuminants = [
    '2HAL_DESK_LED-B025',
    '2HAL_DESK_LED-B050',
    '2HAL_DESK_LED-B075',
    '2HAL_DESK_LED-B100',
    '2HAL_DESK_LED-BG025',
    '2HAL_DESK_LED-BG050',
    '2HAL_DESK_LED-BG075',
    '2HAL_DESK_LED-BG100',
    '2HAL_DESK_R025',
    '2HAL_DESK_R050',
    '2HAL_DESK_R075',
    '2HAL_DESK_R100',
    '2HAL_DESK_RG025',
    '2HAL_DESK_RG050',
    '2HAL_DESK_RG075',
    '2HAL_DESK_RG100',
    '2HAL_DESK',
    '2HAL']


def img_key(scene_id, crop_label, illuminant):
    return f'{scene_id}-{crop_label}-{illuminant}'


def create_dataset():
    """
    Dataset as a dict. Keys are of the form: 04-topleft-2HAL_DESK_LED-B025.
    Images are 0-1 tensors.
    """
    images = dict()
    for s in scenes:
        for ill in illuminants:
            img_path = ds_path / 'images_preview' / f'{s:02d}' / f'{s:02d}_{ill}.jpg'
            # img is loaded as a [0-255] tensor with shape [3, 224, 224], so CHW.
            img = tv.io.read_image(str(img_path)) / 255.0
            cropped_images = pre_norm_transform(img)
            for label, ci in zip(crops, cropped_images):
                images[img_key(s, label, ill)] = ci
    return images

def entropy(x):
    x = x / x.sum()
    entr = -np.sum(x*np.log2(x))
    return entr
diversity_measure = entropy

def test_model():
    """
    (scene, illuminant, crop): label, pred, pred_val
    """
    ds = create_dataset()
    diversity_data = dict()
    cell_size = 2
    raw_data = np.zeros((len(scenes), len(crops), len(illuminants), cell_size))
    classification_val_id_pairs = []
    for s_idx, s in enumerate(scenes): 
        for crop_idx, crop in enumerate(crops):
            classification_ids = []
            for ill_idx, ill in enumerate(illuminants):
                img = ds[img_key(s, crop, ill)]
                input_ = torch.unsqueeze(normalize_transform(img), 0).to(device)
                model_out = model.forward(input_)
                mval, class_id = torch.max(model_out, dim=1)
                classification_val_id_pairs.append((float(mval.cpu()), int(class_id)))
                classification_ids.append(class_id)
                raw_data[s_idx][crop_idx][ill_idx] = (class_id.cpu(), mval.cpu())
            _, unique_counts = np.unique(classification_ids, return_counts=True)
            diversity_data[f'{s}-{crop}'] = diversity_measure(unique_counts)
    xdata = xr.DataArray(raw_data, coords={'scene':scenes, 'crop':crops, 
                                           'illuminant':illuminants, 
                                           'data':['class_id', 'mval']})
    return diversity_data, xdata, classification_val_id_pairs
plot_data, data, val_id_pairs = test_model()

pd.set_option('display.max_rows', None)  
pd.Series(plot_data).to_frame()
#data.values.shape
fd = data.values.reshape((-1, 18, 2))
diversities = []
ave_scores = []
for sample in fd:
    _, unique_counts = np.unique(sample[:,0], return_counts=True)
    d = diversity_measure(unique_counts)
    s = np.mean(sample[:,1])
    diversities.append(d)
    ave_scores.append(s)
plt.scatter(diversities, ave_scores)
plt.title('Classification confidence vs classification diversity')
plt.xlabel('Diversity (entropy, base 2)')
plt.ylabel('Confidence (Max output activation of ResNet)')

Text(0, 0.5, 'Confidence (Max output activation of ResNet)')

png

mean_diversity = np.array(diversities).mean()
mean_diversity

1.3674676955645082

This seems higher than I was expecting. Let’s check out the actual classifications. We don’t have ground truth, but looking at the original 24 images, we have an idea of what is possible. Below is printed out the classifications and the confidence scores (just the max activation of the output layer).

print('Classification:\n(max activation, class label)')
for val, idx in val_id_pairs:
    print(f'{val:.3f},\t{class_id_to_label(idx)}')

Classification:
(max activation, class label)
8.198,	shower curtain
8.723,	shower curtain
9.547,	spotlight
11.780,	spotlight
8.620,	shower curtain
7.484,	shower curtain
7.743,	envelope
6.582,	bath towel
9.759,	shower curtain
9.255,	shower curtain
8.876,	T-shirt
7.568,	shower curtain
8.969,	shower curtain
8.168,	shower curtain
7.053,	T-shirt
6.139,	shower curtain
9.445,	shower curtain
5.306,	lampshade
6.314,	shower curtain
6.337,	shower curtain
6.342,	spotlight
9.483,	spotlight
7.638,	shower curtain
8.864,	shower curtain
7.108,	shower curtain
7.264,	shower curtain
6.646,	lampshade
6.674,	shower curtain
6.671,	shower curtain
6.843,	shower curtain
6.891,	shower curtain
7.206,	shower curtain
7.704,	toilet paper
8.199,	toilet paper
7.284,	shower curtain
5.509,	lampshade
10.223,	envelope
9.441,	envelope
11.627,	spotlight
11.799,	spotlight
11.354,	envelope
11.258,	envelope
10.813,	envelope
10.197,	envelope
7.339,	ring binder
6.866,	envelope
7.437,	lampshade
8.409,	couch
8.046,	envelope
8.294,	envelope
7.970,	envelope
9.072,	envelope
7.535,	envelope
6.967,	envelope
10.975,	speaker
12.401,	home theater
13.418,	speaker
13.630,	speaker
8.850,	carton
10.026,	carton
10.283,	carton
9.609,	carton
10.697,	home theater
11.868,	speaker
12.378,	speaker
13.873,	speaker
10.183,	speaker
9.109,	speaker
9.690,	carton
10.184,	perfume
10.466,	speaker
9.091,	speaker
10.001,	home theater
9.092,	home theater
11.922,	stage
12.202,	stage
8.616,	home theater
8.121,	home theater
8.258,	home theater
8.823,	home theater
9.591,	home theater
9.763,	home theater
9.393,	home theater
10.047,	home theater
9.230,	home theater
7.640,	home theater
7.862,	baluster
7.798,	baluster
9.556,	home theater
6.630,	home theater
6.665,	shower curtain
7.215,	T-shirt
9.448,	spotlight
11.268,	spotlight
7.546,	shower curtain
6.326,	shower curtain
6.306,	shower curtain
5.889,	envelope
8.481,	shower curtain
7.581,	shower curtain
7.604,	T-shirt
6.334,	shower curtain
7.322,	shower curtain
6.789,	mosquito net
7.187,	envelope
5.267,	shower curtain
8.164,	shower curtain
5.418,	spotlight
6.426,	shower curtain
6.242,	shower curtain
6.278,	front curtain
9.257,	spotlight
8.418,	shower curtain
9.239,	shower curtain
7.718,	shower curtain
7.620,	shower curtain
7.014,	shower curtain
6.716,	shower curtain
7.370,	shower curtain
8.083,	shower curtain
6.865,	shower curtain
6.930,	shower curtain
7.137,	shower curtain
8.006,	toilet paper
6.692,	lampshade
6.016,	spotlight
9.682,	envelope
8.408,	envelope
9.509,	spotlight
12.376,	spotlight
10.035,	envelope
11.666,	envelope
11.175,	envelope
10.398,	envelope
8.218,	envelope
8.105,	envelope
7.181,	envelope
6.983,	envelope
9.102,	envelope
9.437,	envelope
9.150,	envelope
10.231,	eraser
8.293,	envelope
6.740,	envelope
10.731,	home theater
11.365,	home theater
13.211,	home theater
14.637,	home theater
8.392,	home theater
8.274,	ring binder
9.028,	ring binder
9.814,	ring binder
11.614,	home theater
11.521,	home theater
12.623,	home theater
13.368,	home theater
9.321,	home theater
8.804,	home theater
8.740,	ring binder
9.798,	ring binder
10.155,	home theater
9.068,	home theater
7.079,	birdhouse
7.595,	picket fence
8.488,	stage
10.616,	stage
9.110,	birdhouse
9.582,	birdhouse
9.808,	birdhouse
8.641,	birdhouse
7.319,	birdhouse
7.285,	picket fence
7.617,	picket fence
7.368,	picket fence
8.482,	birdhouse
8.894,	birdhouse
9.918,	birdhouse
9.907,	birdhouse
6.985,	birdhouse
5.888,	birdhouse
7.477,	shower curtain
7.574,	shower curtain
9.848,	spotlight
11.361,	spotlight
7.325,	shower curtain
7.256,	shower curtain
6.649,	shower curtain
6.626,	bath towel
9.077,	shower curtain
8.980,	shower curtain
7.946,	T-shirt
7.635,	shower curtain
8.663,	shower curtain
7.866,	shower curtain
6.931,	velvet
5.835,	shower curtain
8.731,	shower curtain
5.495,	spotlight
5.659,	shower curtain
6.156,	front curtain
6.993,	front curtain
9.480,	spotlight
7.805,	shower curtain
8.992,	shower curtain
7.793,	shower curtain
7.505,	shower curtain
6.889,	shower curtain
6.765,	shower curtain
7.142,	shower curtain
7.680,	shower curtain
6.960,	shower curtain
7.335,	shower curtain
8.234,	shower curtain
8.869,	toilet paper
7.172,	shower curtain
5.849,	spotlight
11.696,	candle
11.674,	candle
11.560,	digital clock
12.012,	digital clock
12.366,	eraser
14.139,	eraser
13.454,	eraser
13.311,	eraser
11.538,	candle
12.211,	candle
11.145,	candle
11.065,	candle
10.984,	candle
11.969,	eraser
12.537,	eraser
14.300,	eraser
11.424,	candle
7.767,	candle
9.874,	chest
9.435,	chest
10.684,	carton
11.577,	carton
9.710,	carton
9.873,	carton
11.344,	carton
11.795,	carton
9.456,	chest
9.978,	chest
10.162,	couch
10.803,	couch
9.953,	chest
10.139,	carton
10.711,	carton
10.991,	carton
9.411,	carton
9.227,	carton
9.196,	barrel
9.734,	barrel
11.198,	barrel
11.412,	barrel
9.813,	barrel
10.541,	barrel
10.625,	barrel
11.972,	bucket
10.650,	pill bottle
10.110,	pill bottle
12.805,	pill bottle
11.667,	pill bottle
9.626,	barrel
10.342,	barrel
11.577,	barrel
11.019,	barrel
10.213,	pill bottle
11.286,	pill bottle
7.608,	shower curtain
7.951,	shower curtain
9.733,	spotlight
11.422,	spotlight
8.090,	shower curtain
7.252,	shower curtain
7.107,	shower curtain
6.214,	bath towel
9.192,	shower curtain
8.338,	shower curtain
7.498,	shower curtain
7.594,	shower curtain
8.068,	shower curtain
7.504,	shower curtain
6.924,	T-shirt
5.868,	shower curtain
9.197,	shower curtain
5.517,	spotlight
6.404,	shower curtain
6.452,	front curtain
6.447,	front curtain
9.468,	spotlight
7.933,	shower curtain
8.735,	shower curtain
7.881,	shower curtain
7.960,	shower curtain
6.950,	shower curtain
6.825,	lampshade
7.110,	shower curtain
6.972,	shower curtain
6.663,	shower curtain
7.285,	shower curtain
8.346,	toilet paper
9.083,	toilet paper
6.957,	lampshade
5.977,	spotlight
11.031,	abacus
10.535,	dumbbell
12.665,	coil
12.730,	coil
10.610,	dumbbell
10.427,	abacus
9.681,	knot
9.851,	knot
10.860,	dumbbell
11.023,	abacus
11.716,	coil
12.380,	coil
10.686,	dumbbell
9.550,	dumbbell
9.989,	coil
10.807,	bell pepper
11.112,	dumbbell
7.718,	eraser
13.762,	basketball
13.378,	basketball
13.497,	basketball
13.279,	basketball
15.574,	basketball
15.702,	basketball
15.288,	basketball
15.742,	basketball
12.564,	jack-o'-lantern
12.592,	jack-o'-lantern
12.122,	jack-o'-lantern
11.198,	basketball
14.337,	basketball
15.082,	basketball
14.558,	basketball
13.839,	basketball
13.625,	basketball
12.475,	basketball
8.428,	front curtain
9.310,	front curtain
11.075,	stage
11.453,	stage
7.934,	table lamp
7.488,	table lamp
7.894,	baluster
8.022,	home theater
8.330,	table lamp
8.202,	table lamp
7.725,	table lamp
8.531,	home theater
8.021,	table lamp
7.973,	baluster
8.811,	baluster
8.500,	baluster
8.093,	table lamp
6.360,	lampshade
6.522,	front curtain
6.665,	front curtain
6.822,	front curtain
6.709,	front curtain
6.580,	front curtain
6.543,	front curtain
7.032,	front curtain
7.104,	front curtain
7.108,	front curtain
7.129,	front curtain
6.849,	apron
7.683,	apron
6.940,	front curtain
7.329,	front curtain
6.860,	front curtain
7.219,	apron
5.750,	front curtain
5.410,	match
5.264,	Granny Smith
5.222,	Granny Smith
5.369,	spotlight
5.337,	microphone
6.429,	Granny Smith
7.234,	Granny Smith
6.910,	Granny Smith
6.939,	Granny Smith
5.248,	Granny Smith
5.075,	Granny Smith
5.203,	jack-o'-lantern
5.274,	jack-o'-lantern
6.902,	Granny Smith
7.084,	Granny Smith
7.971,	Granny Smith
8.257,	Granny Smith
5.484,	Granny Smith
5.331,	tennis ball
8.293,	digital clock
8.490,	digital clock
8.421,	digital clock
8.147,	digital clock
8.392,	digital clock
9.746,	digital clock
10.454,	digital clock
10.819,	digital clock
8.280,	digital clock
8.280,	digital clock
8.752,	digital clock
8.836,	digital clock
8.236,	digital clock
8.906,	digital clock
10.009,	digital clock
10.542,	digital clock
7.979,	digital clock
6.369,	lighter
8.824,	pill bottle
8.642,	pill bottle
7.835,	pill bottle
7.837,	billiard table
8.244,	pill bottle
7.708,	pill bottle
7.622,	pill bottle
7.747,	pill bottle
9.963,	pill bottle
10.147,	pill bottle
10.284,	pill bottle
10.080,	pill bottle
9.123,	pill bottle
9.367,	pill bottle
8.112,	pill bottle
7.738,	pill bottle
9.100,	pill bottle
8.642,	pill bottle
6.956,	balloon
6.816,	billiard table
7.459,	billiard table
7.860,	billiard table
8.593,	Granny Smith
9.347,	Granny Smith
9.586,	billiard table
9.327,	billiard table
6.939,	Granny Smith
6.956,	pill bottle
7.541,	joystick
9.158,	billiard table
8.660,	Granny Smith
9.154,	Granny Smith
10.565,	Granny Smith
10.430,	Granny Smith
7.493,	Granny Smith
5.951,	spotlight
5.843,	spotlight
6.329,	spotlight
6.710,	spotlight
7.176,	spotlight
6.051,	spotlight
6.075,	spotlight
6.477,	spotlight
5.926,	stage
5.702,	match
5.610,	match
5.752,	spotlight
5.841,	spotlight
5.558,	match
5.079,	spotlight
5.825,	spotlight
5.843,	window shade
5.741,	match
4.296,	match
8.338,	spotlight
9.527,	balloon
8.843,	spotlight
8.671,	spotlight
7.100,	billiard table
7.874,	balloon
8.630,	spotlight
9.412,	bubble
6.953,	spotlight
7.334,	balloon
8.073,	orange
7.804,	balloon
8.372,	orange
9.332,	orange
11.800,	orange
13.032,	orange
6.709,	orange
6.493,	spotlight
9.945,	mortar
9.413,	mortar
9.913,	candle
11.255,	candle
8.447,	mortar
8.229,	mortar
9.584,	Granny Smith
10.366,	Granny Smith
10.000,	mortar
9.712,	mortar
9.256,	mortar
8.989,	mortar
9.608,	mortar
9.197,	mortar
10.163,	bagel
8.697,	bagel
10.083,	mortar
10.104,	mortar
10.247,	ping-pong ball
9.445,	match
9.546,	match
9.856,	match
10.363,	ping-pong ball
9.790,	ping-pong ball
8.622,	dumbbell
8.181,	dumbbell
10.554,	match
10.334,	candle
9.889,	ping-pong ball
9.450,	candle
10.421,	candle
10.669,	candle
10.829,	candle
11.080,	candle
10.611,	match
8.936,	ping-pong ball
9.431,	orange
8.930,	balloon
9.369,	spotlight
9.598,	spotlight
9.351,	orange
8.442,	balloon
8.614,	balloon
8.790,	ping-pong ball
10.422,	orange
9.791,	orange
10.083,	jack-o'-lantern
9.928,	orange
11.419,	orange
11.547,	orange
11.725,	orange
10.984,	orange
11.424,	orange
6.871,	jack-o'-lantern
5.799,	match
5.602,	match
6.085,	spotlight
6.823,	spotlight
5.875,	spotlight
5.837,	spotlight
6.536,	spotlight
6.552,	stage
6.004,	match
5.745,	match
5.816,	match
5.555,	spotlight
5.373,	match
5.274,	spotlight
5.847,	spotlight
5.739,	lampshade
5.822,	match
4.326,	match
6.101,	match
4.939,	match
5.701,	match
5.764,	spotlight
5.283,	match
6.187,	match
6.147,	spotlight
5.983,	spotlight
5.551,	match
5.093,	match
5.274,	match
5.412,	match
4.918,	match
5.363,	match
5.410,	match
5.706,	spotlight
5.562,	match
4.427,	nematode
8.829,	jack-o'-lantern
8.272,	jack-o'-lantern
8.302,	jellyfish
8.787,	jellyfish
8.601,	jack-o'-lantern
8.051,	jack-o'-lantern
8.707,	tennis ball
10.119,	tennis ball
8.894,	jack-o'-lantern
9.184,	jack-o'-lantern
9.221,	jack-o'-lantern
8.708,	jack-o'-lantern
8.767,	jack-o'-lantern
8.499,	jack-o'-lantern
7.410,	jack-o'-lantern
8.290,	wool
8.542,	jack-o'-lantern
8.137,	jack-o'-lantern
10.324,	eraser
9.547,	eraser
7.442,	match
7.290,	digital clock
10.692,	eraser
9.939,	eraser
6.925,	sock
7.199,	sock
11.278,	eraser
11.754,	eraser
10.793,	eraser
10.317,	eraser
11.417,	eraser
11.324,	eraser
9.699,	eraser
10.279,	eraser
10.698,	eraser
10.451,	eraser
7.414,	lampshade
7.279,	lampshade
6.416,	jack-o'-lantern
7.231,	basketball
6.540,	Band-Aid
7.587,	eraser
8.325,	tennis ball
9.214,	tennis ball
7.148,	lampshade
6.690,	lampshade
7.344,	lampshade
7.438,	lampshade
6.849,	Band-Aid
7.497,	tennis ball
9.363,	eraser
10.307,	eraser
7.215,	lampshade
7.328,	lampshade
5.778,	front curtain
6.380,	front curtain
6.720,	front curtain
6.609,	front curtain
6.478,	front curtain
6.505,	front curtain
6.555,	front curtain
6.225,	front curtain
6.066,	front curtain
6.499,	front curtain
6.120,	front curtain
6.382,	front curtain
6.617,	front curtain
6.314,	front curtain
6.194,	front curtain
6.158,	front curtain
5.918,	front curtain
5.382,	match
5.711,	match
5.669,	match
5.631,	match
5.595,	match
5.612,	match
5.671,	match
5.678,	match
5.533,	spotlight
5.641,	match
5.544,	match
5.110,	spotlight
4.690,	front curtain
5.890,	match
5.455,	match
5.191,	spotlight
4.794,	front curtain
5.688,	match
4.819,	match
8.357,	home theater
8.783,	digital clock
11.206,	digital clock
11.978,	digital clock
8.811,	home theater
8.576,	home theater
7.590,	home theater
7.961,	digital clock
7.330,	home theater
7.303,	home theater
7.484,	home theater
7.325,	home theater
7.271,	home theater
7.454,	home theater
7.660,	digital clock
7.905,	digital clock
7.178,	orange
5.844,	jack-o'-lantern
9.488,	Band-Aid
8.906,	Band-Aid
10.708,	Band-Aid
10.464,	Band-Aid
9.790,	Band-Aid
9.706,	Band-Aid
10.741,	Band-Aid
10.452,	Band-Aid
9.281,	Band-Aid
9.769,	Band-Aid
9.836,	Band-Aid
9.635,	Band-Aid
10.306,	Band-Aid
9.835,	Band-Aid
11.773,	Band-Aid
12.051,	Band-Aid
10.488,	Band-Aid
10.121,	Band-Aid
13.124,	Band-Aid
13.681,	Band-Aid
12.223,	Band-Aid
12.013,	eraser
11.801,	Band-Aid
10.921,	Band-Aid
10.957,	eraser
11.370,	eraser
15.842,	Band-Aid
15.816,	Band-Aid
15.927,	Band-Aid
15.470,	Band-Aid
15.009,	Band-Aid
14.721,	Band-Aid
14.221,	Band-Aid
12.202,	Band-Aid
14.820,	Band-Aid
11.317,	Band-Aid
6.277,	front curtain
6.128,	front curtain
6.238,	front curtain
6.538,	front curtain
5.981,	front curtain
5.995,	front curtain
7.047,	front curtain
7.585,	front curtain
6.535,	front curtain
7.310,	front curtain
6.493,	front curtain
7.514,	apron
6.423,	front curtain
6.502,	front curtain
6.327,	apron
6.986,	apron
6.169,	front curtain
5.266,	match
5.690,	spotlight
5.045,	front curtain
6.214,	jellyfish
7.038,	jellyfish
4.968,	spotlight
5.317,	teddy bear
5.821,	teddy bear
5.191,	teddy bear
4.755,	front curtain
5.681,	front curtain
6.602,	front curtain
7.080,	front curtain
4.639,	spotlight
5.134,	teddy bear
5.785,	teddy bear
5.716,	teddy bear
5.262,	spotlight
5.129,	match
8.023,	wool
8.030,	wool
7.600,	wool
7.978,	wool
9.610,	wool
10.541,	wool
10.437,	wool
11.125,	wool
8.553,	wool
8.761,	wool
9.215,	wool
9.984,	wool
10.071,	wool
10.872,	wool
10.627,	wool
9.892,	wool
8.158,	wool
5.689,	jack-o'-lantern
8.697,	teddy bear
8.844,	teddy bear
8.754,	teddy bear
8.083,	front curtain
7.419,	teddy bear
6.865,	Pembroke Welsh Corgi
7.918,	tennis ball
6.942,	tennis ball
9.915,	teddy bear
10.171,	teddy bear
9.519,	teddy bear
9.655,	teddy bear
9.392,	teddy bear
8.328,	teddy bear
8.529,	Pembroke Welsh Corgi
8.556,	teddy bear
9.251,	teddy bear
8.924,	teddy bear
9.629,	teddy bear
10.427,	clownfish
9.071,	clownfish
10.191,	stage
10.514,	teddy bear
9.322,	teddy bear
8.898,	flamingo
8.329,	flamingo
11.236,	teddy bear
10.542,	teddy bear
10.102,	teddy bear
9.680,	teddy bear
10.819,	teddy bear
10.112,	teddy bear
9.972,	teddy bear
10.499,	teddy bear
10.671,	teddy bear
7.690,	teddy bear
5.853,	spotlight
5.981,	spotlight
6.586,	spotlight
7.218,	spotlight
6.265,	spotlight
6.001,	spotlight
6.469,	spotlight
5.926,	spotlight
5.701,	match
5.699,	spotlight
5.879,	spotlight
5.867,	spotlight
5.521,	match
5.147,	spotlight
5.777,	spotlight
6.060,	spotlight
5.820,	match
4.297,	match
5.574,	match
5.047,	match
5.978,	match
6.164,	match
5.235,	match
6.225,	match
6.077,	match
6.580,	match
5.048,	match
5.211,	match
5.518,	match
5.541,	match
5.174,	match
5.369,	match
5.394,	match
5.233,	spotlight
5.293,	match
4.471,	nematode
7.122,	bath towel
6.503,	bath towel
6.405,	front curtain
6.591,	sock
6.881,	bath towel
7.127,	bath towel
7.722,	bath towel
8.371,	bath towel
7.820,	bath towel
8.052,	bath towel
7.688,	sleeping bag
8.319,	sweatshirt
7.855,	bath towel
8.130,	bath towel
9.235,	bath towel
8.959,	bath towel
7.337,	bath towel
6.139,	candle
11.681,	bath towel
9.036,	bath towel
8.501,	bath towel
9.130,	bath towel
12.122,	bath towel
10.105,	bath towel
12.337,	bath towel
14.073,	bath towel
12.122,	bath towel
11.882,	bath towel
11.793,	bath towel
11.031,	bath towel
12.829,	bath towel
13.381,	bath towel
15.450,	bath towel
14.675,	bath towel
11.873,	bath towel
8.548,	mosquito net
6.607,	quilt
6.112,	velvet
6.010,	velvet
5.583,	velvet
6.526,	quilt
6.491,	quilt
8.029,	bath towel
8.219,	bath towel
6.929,	quilt
6.798,	quilt
7.940,	quilt
8.079,	quilt
6.773,	quilt
7.126,	quilt
8.231,	bath towel
8.237,	bath towel
6.748,	quilt
5.761,	sleeping bag
5.820,	match
6.110,	spotlight
6.436,	spotlight
6.671,	spotlight
6.037,	spotlight
6.111,	spotlight
6.213,	spotlight
6.035,	jellyfish
5.975,	match
5.781,	match
5.653,	spotlight
5.586,	spotlight
5.670,	match
5.318,	spotlight
5.511,	spotlight
5.111,	spotlight
5.820,	match
4.397,	match
5.654,	match
5.299,	match
5.741,	match
5.851,	match
5.060,	match
6.296,	match
6.171,	match
6.407,	spotlight
5.248,	match
5.204,	match
5.380,	match
5.882,	match
5.054,	match
5.405,	match
5.556,	match
5.727,	spotlight
5.100,	match
4.438,	nematode
12.551,	spatula
12.509,	spatula
12.956,	spatula
13.726,	spatula
10.927,	spatula
12.673,	spatula
12.661,	spatula
13.510,	spatula
11.968,	spatula
12.306,	spatula
13.326,	spatula
14.191,	spatula
12.049,	spatula
13.550,	spatula
15.003,	spatula
14.845,	spatula
11.602,	spatula
10.655,	spatula
7.777,	cradle
6.899,	cradle
6.756,	wool
7.346,	match
7.906,	bath towel
7.675,	Chesapeake Bay Retriever
7.740,	Chesapeake Bay Retriever
6.654,	Chesapeake Bay Retriever
8.116,	bath towel
7.818,	bath towel
7.853,	orange
7.676,	orange
8.186,	bath towel
8.407,	bath towel
7.898,	bath towel
9.032,	lemon
7.436,	bath towel
7.407,	bath towel
6.612,	Band-Aid
6.246,	Band-Aid
5.706,	microphone
5.889,	hourglass
5.949,	hourglass
7.171,	Band-Aid
6.744,	Band-Aid
7.303,	tennis ball
6.019,	Band-Aid
6.729,	Band-Aid
6.215,	hourglass
6.610,	dumbbell
7.119,	Band-Aid
7.439,	Band-Aid
8.383,	Band-Aid
8.469,	orange
6.229,	Band-Aid
5.163,	Band-Aid
8.548,	acoustic guitar
8.497,	acoustic guitar
8.345,	paintbrush
9.902,	paintbrush
9.444,	acoustic guitar
9.089,	acoustic guitar
9.020,	toilet seat
8.341,	broom
8.665,	acoustic guitar
9.102,	acoustic guitar
10.059,	acoustic guitar
9.188,	acoustic guitar
8.945,	acoustic guitar
8.824,	acoustic guitar
8.789,	acoustic guitar
8.066,	acoustic guitar
8.271,	acoustic guitar
6.924,	lampshade
5.339,	front curtain
6.432,	front curtain
6.447,	front curtain
6.648,	front curtain
5.299,	front curtain
4.838,	front curtain
4.617,	front curtain
4.290,	front curtain
6.072,	front curtain
6.742,	front curtain
9.175,	front curtain
10.604,	front curtain
5.374,	front curtain
5.931,	front curtain
6.880,	front curtain
7.750,	front curtain
5.732,	front curtain
4.365,	match
10.918,	cup
10.934,	cup
11.336,	cup
11.516,	cup
10.725,	cup
10.173,	coffee mug
9.818,	coffee mug
10.198,	coffee mug
11.196,	cup
10.779,	cup
10.444,	cup
10.296,	cup
10.784,	cup
10.837,	cup
10.016,	cup
9.920,	cup
11.414,	cup
8.674,	cup
10.187,	candle
9.392,	candle
9.476,	candle
9.096,	candle
11.239,	candle
10.145,	candle
9.292,	ice pop
10.632,	ice pop
10.729,	candle
9.908,	candle
9.937,	candle
9.331,	couch
11.776,	candle
10.488,	candle
10.074,	candle
9.599,	ice pop
11.325,	candle
6.976,	candle
9.483,	wooden spoon
8.656,	wooden spoon
9.264,	soup bowl
9.179,	soup bowl
8.625,	plate rack
9.269,	plate rack
8.400,	soup bowl
8.586,	chocolate syrup
9.474,	wooden spoon
9.895,	wooden spoon
10.505,	measuring cup
11.534,	wooden spoon
9.007,	wooden spoon
8.820,	mixing bowl
9.253,	mixing bowl
9.289,	mixing bowl
9.010,	wooden spoon
8.261,	lampshade
6.296,	match
6.225,	match
6.901,	accordion
7.825,	accordion
6.480,	match
6.229,	match
5.732,	match
5.292,	match
7.057,	match
7.359,	match
7.519,	match
6.774,	nail
6.714,	match
6.770,	match
6.193,	match
6.431,	leafhopper
6.955,	match
4.021,	daisy
6.114,	match
5.256,	match
5.411,	match
5.906,	spotlight
5.272,	match
6.022,	match
5.764,	match
6.008,	spotlight
5.397,	match
5.447,	match
5.128,	match
5.652,	match
4.955,	match
5.322,	match
5.683,	match
5.579,	spotlight
5.591,	match
4.526,	nematode
11.588,	bucket
10.940,	bucket
10.777,	bucket
10.585,	bucket
12.032,	bucket
11.981,	bucket
10.820,	bucket
9.717,	bucket
11.837,	bucket
12.009,	bucket
12.332,	bucket
12.635,	bucket
12.118,	bucket
12.409,	bucket
12.899,	bucket
12.649,	bucket
11.636,	bucket
4.996,	pill bottle
10.007,	jellyfish
9.487,	jellyfish
9.755,	jellyfish
9.085,	jellyfish
9.554,	jellyfish
8.852,	hourglass
9.772,	piggy bank
11.015,	piggy bank
9.726,	jellyfish
9.624,	jellyfish
10.460,	candle
11.639,	candle
8.924,	jellyfish
8.845,	jellyfish
9.531,	candle
9.999,	piggy bank
10.130,	jellyfish
10.843,	jellyfish
8.556,	candle
8.885,	candle
9.135,	candle
8.915,	pencil sharpener
9.112,	candle
8.786,	candle
8.590,	candle
9.277,	pencil sharpener
9.341,	candle
9.434,	candle
9.380,	candle
8.865,	candle
8.943,	candle
8.905,	candle
7.756,	candle
7.434,	cucumber
8.981,	candle
6.254,	Band-Aid
6.169,	spotlight
6.028,	spotlight
6.127,	spotlight
6.843,	spotlight
6.326,	spotlight
5.823,	spotlight
6.534,	spotlight
6.530,	stage
5.924,	match
5.839,	match
5.847,	match
5.784,	spotlight
5.613,	match
5.707,	spotlight
6.444,	spotlight
5.581,	spotlight
5.919,	match
4.318,	match
5.980,	spotlight
4.749,	match
6.296,	spotlight
5.715,	match
4.790,	match
6.041,	spotlight
6.118,	spotlight
6.334,	spotlight
5.231,	match
5.539,	match
4.809,	match
5.389,	match
4.792,	match
5.517,	match
5.970,	spotlight
5.748,	spotlight
5.660,	match
4.454,	nematode
7.286,	match
7.299,	match
8.808,	match
8.497,	jellyfish
8.089,	lemon
8.893,	Granny Smith
10.820,	Granny Smith
11.569,	Granny Smith
8.059,	match
8.343,	match
9.049,	match
9.850,	match
8.633,	lemon
9.471,	Granny Smith
11.837,	Granny Smith
13.248,	Granny Smith
7.708,	match
5.753,	lemon
6.593,	ping-pong ball
6.377,	ping-pong ball
7.520,	jack-o'-lantern
6.977,	spotlight
6.602,	Granny Smith
8.099,	Granny Smith
9.734,	Granny Smith
9.846,	Granny Smith
7.926,	ping-pong ball
7.292,	ping-pong ball
7.104,	orange
7.837,	orange
6.799,	ping-pong ball
7.571,	Granny Smith
8.278,	Granny Smith
10.176,	Granny Smith
6.784,	ping-pong ball
5.648,	tennis ball
9.014,	match
8.972,	lighter
9.415,	match
9.181,	match
11.151,	Granny Smith
14.051,	Granny Smith
15.755,	Granny Smith
14.660,	Granny Smith
10.299,	orange
11.238,	orange
12.000,	orange
12.349,	orange
11.539,	lemon
12.529,	Granny Smith
15.397,	Granny Smith
16.749,	Granny Smith
10.214,	orange
5.533,	match
7.318,	front curtain
7.513,	front curtain
11.317,	spotlight
10.899,	spotlight
7.754,	shower curtain
7.810,	mosquito net
8.429,	envelope
6.314,	shower curtain
8.528,	shower curtain
7.536,	shower curtain
7.326,	shower curtain
6.711,	shower curtain
7.348,	shower curtain
6.565,	shower curtain
7.754,	lampshade
6.917,	shower curtain
7.254,	shower curtain
5.456,	spotlight
4.435,	velvet
5.724,	velvet
7.531,	spotlight
10.322,	spotlight
7.152,	shower curtain
8.051,	shower curtain
7.679,	envelope
7.373,	envelope
6.314,	shower curtain
5.927,	toilet paper
6.075,	toilet paper
6.179,	toilet paper
6.910,	toilet paper
7.274,	toilet paper
9.076,	toilet paper
8.635,	envelope
6.585,	lampshade
5.858,	spotlight
7.033,	orange
7.520,	candle
10.527,	mortar
13.500,	mortar
7.784,	mortar
8.089,	bell pepper
10.359,	cucumber
11.065,	bell pepper
7.141,	mixing bowl
7.167,	orange
7.867,	bell pepper
7.903,	bell pepper
6.995,	bell pepper
7.929,	bell pepper
11.296,	cucumber
11.674,	cucumber
7.402,	mortar
7.001,	candle
8.927,	bell pepper
9.566,	maraca
9.970,	maraca
10.435,	jellyfish
11.516,	bell pepper
11.824,	bell pepper
13.826,	bell pepper
15.318,	bell pepper
11.886,	bell pepper
11.240,	bell pepper
10.220,	bell pepper
9.501,	bell pepper
13.354,	bell pepper
13.326,	bell pepper
14.356,	bell pepper
15.081,	bell pepper
11.590,	bell pepper
9.401,	bell pepper
10.725,	bell pepper
11.909,	bell pepper
11.084,	jellyfish
11.747,	jellyfish
15.122,	bell pepper
15.539,	bell pepper
16.185,	bell pepper
14.673,	bell pepper
13.562,	bell pepper
13.988,	bell pepper
14.172,	bell pepper
12.459,	bell pepper
14.930,	bell pepper
16.662,	bell pepper
15.994,	bell pepper
16.250,	bell pepper
12.802,	bell pepper
8.222,	candle
6.834,	front curtain
7.468,	front curtain
7.828,	front curtain
8.595,	front curtain
7.434,	front curtain
6.993,	front curtain
7.673,	front curtain
7.153,	front curtain
8.008,	front curtain
7.997,	front curtain
8.231,	front curtain
8.137,	front curtain
7.799,	front curtain
7.980,	front curtain
7.865,	front curtain
7.773,	front curtain
7.055,	front curtain
5.549,	match
5.918,	match
5.914,	match
5.780,	match
5.672,	spotlight
5.892,	match
5.995,	match
5.878,	match
5.942,	spotlight
6.021,	match
5.890,	spotlight
5.449,	spotlight
5.965,	front curtain
5.964,	match
5.634,	match
5.666,	spotlight
5.733,	front curtain
5.894,	spotlight
4.787,	match
14.876,	banana
15.481,	banana
14.528,	banana
13.928,	banana
14.932,	banana
15.417,	banana
15.016,	banana
14.225,	banana
14.983,	banana
14.712,	banana
12.824,	banana
12.043,	banana
15.434,	banana
15.424,	banana
14.951,	banana
15.003,	banana
14.673,	banana
13.179,	banana
9.891,	pill bottle
9.556,	pill bottle
9.008,	pill bottle
9.411,	jellyfish
9.199,	pill bottle
8.463,	pill bottle
7.941,	lemon
8.467,	lemon
10.909,	pill bottle
11.544,	pill bottle
11.394,	pill bottle
11.542,	pill bottle
9.151,	pill bottle
8.561,	pill bottle
8.877,	lemon
9.150,	lemon
10.631,	pill bottle
9.209,	pill bottle
11.080,	orange
10.643,	orange
11.053,	orange
11.032,	orange
12.434,	orange
13.663,	orange
14.868,	orange
13.202,	orange
11.229,	orange
11.595,	orange
11.153,	orange
10.285,	orange
11.929,	orange
12.876,	orange
13.910,	orange
13.796,	orange
11.790,	orange
6.463,	pill bottle
5.732,	front curtain
6.053,	front curtain
6.005,	front curtain
7.445,	front curtain
6.073,	front curtain
6.377,	front curtain
6.577,	front curtain
6.408,	front curtain
6.053,	front curtain
6.572,	front curtain
7.381,	front curtain
8.122,	front curtain
6.007,	front curtain
6.479,	front curtain
7.114,	front curtain
7.890,	front curtain
6.162,	front curtain
5.528,	match
6.282,	lighter
6.534,	lighter
6.308,	spotlight
6.813,	spotlight
6.673,	lighter
7.534,	lighter
7.003,	lighter
7.067,	lighter
6.895,	lighter
7.086,	lighter
6.208,	lighter
6.123,	spotlight
7.109,	lighter
6.924,	lighter
6.528,	perfume
6.442,	spotlight
6.900,	lighter
4.134,	spotlight
7.232,	paintbrush
7.821,	ballpoint pen
8.858,	ballpoint pen
8.958,	eraser
7.742,	paintbrush
7.241,	flute
9.153,	screwdriver
9.078,	screwdriver
6.976,	paintbrush
6.529,	trombone
5.854,	screwdriver
5.730,	syringe
7.023,	paintbrush
8.280,	fountain pen
8.889,	fountain pen
8.513,	ballpoint pen
7.701,	paintbrush
6.005,	safety pin
7.741,	lighter
7.943,	lighter
7.762,	lighter
7.895,	lighter
8.336,	lighter
7.997,	lighter
7.998,	lighter
7.514,	lighter
8.294,	lighter
7.661,	lighter
8.607,	lighter
8.433,	lighter
7.471,	lighter
7.715,	lighter
8.039,	lighter
7.971,	lighter
7.762,	lighter
7.070,	lighter
9.105,	microphone
9.236,	microphone
10.125,	microphone
10.570,	microphone
9.922,	microphone
9.997,	microphone
10.890,	microphone
11.755,	microphone
9.602,	microphone
9.576,	microphone
9.656,	microphone
9.756,	microphone
9.377,	microphone
9.525,	binoculars
9.946,	microphone
10.521,	screwdriver
9.378,	syringe
8.827,	pill bottle
9.052,	shower curtain
11.340,	shower curtain
10.898,	stage
11.332,	spotlight
10.335,	shower curtain
9.601,	shower curtain
9.278,	handkerchief
7.754,	shower curtain
11.119,	shower curtain
11.419,	shower curtain
9.605,	shower curtain
8.397,	shower curtain
12.113,	shower curtain
11.474,	shower curtain
8.177,	apron
5.948,	shower curtain
9.473,	shower curtain
6.371,	spotlight
6.926,	shower curtain
7.774,	apron
6.786,	velvet
8.369,	stage
7.993,	shower curtain
8.219,	shower curtain
7.977,	apron
8.269,	apron
6.149,	lampshade
5.983,	shower curtain
6.382,	apron
6.914,	apron
7.007,	shower curtain
7.226,	shower curtain
8.272,	shower curtain
8.620,	apron
6.363,	lampshade
5.860,	spotlight
9.186,	match
10.093,	match
10.330,	pill bottle
9.963,	pill bottle
9.281,	screwdriver
9.494,	screwdriver
9.451,	screwdriver
10.685,	screwdriver
9.253,	screwdriver
9.620,	screwdriver
10.983,	screwdriver
11.977,	screwdriver
9.213,	screwdriver
9.563,	screwdriver
10.094,	screwdriver
12.373,	match
9.169,	screwdriver
9.775,	match
8.444,	ring binder
8.476,	ring binder
8.905,	home theater
8.443,	digital clock
8.072,	flute
8.872,	ring binder
8.620,	ring binder
8.430,	ring binder
8.073,	ring binder
8.188,	ring binder
8.128,	home theater
7.698,	home theater
7.316,	ring binder
8.338,	flute
8.233,	flute
7.876,	flute
7.721,	ring binder
6.361,	lighter
8.933,	jack-o'-lantern
7.817,	jack-o'-lantern
10.460,	stage
12.329,	stage
8.256,	butternut squash
8.040,	jack-o'-lantern
6.677,	stage
7.603,	stage
8.575,	jack-o'-lantern
9.979,	jack-o'-lantern
12.292,	jack-o'-lantern
12.936,	jack-o'-lantern
7.429,	butternut squash
8.709,	jack-o'-lantern
7.927,	jack-o'-lantern
8.807,	cornet
8.126,	butternut squash
6.797,	butternut squash
5.440,	front curtain
5.345,	front curtain
5.908,	front curtain
6.556,	front curtain
4.720,	front curtain
4.444,	front curtain
5.095,	front curtain
5.944,	front curtain
6.021,	front curtain
6.577,	front curtain
6.773,	quilt
6.608,	apron
4.732,	front curtain
4.695,	bridegroom
6.932,	apron
8.379,	apron
5.357,	front curtain
4.531,	match
5.290,	spotlight
5.217,	spotlight
5.183,	spotlight
5.169,	spotlight
5.422,	spotlight
5.296,	spotlight
5.196,	spotlight
5.148,	spotlight
5.131,	spotlight
5.323,	spotlight
6.405,	front curtain
6.285,	front curtain
5.042,	spotlight
5.230,	front curtain
5.812,	front curtain
6.571,	front curtain
5.191,	match
4.976,	match
11.446,	cup
11.040,	cup
10.925,	cup
11.868,	cup
11.345,	cup
11.581,	teapot
12.813,	teapot
13.331,	teapot
11.634,	teapot
11.824,	cup
11.921,	cup
11.502,	cup
12.063,	cup
12.544,	cup
13.233,	cup
14.109,	cup
11.101,	cup
11.664,	cup
8.086,	pencil sharpener
8.694,	pencil sharpener
8.998,	pencil sharpener
9.193,	pencil sharpener
7.641,	lampshade
7.861,	thimble
7.887,	drum
7.774,	pencil sharpener
7.326,	lampshade
7.309,	lampshade
8.503,	drum
8.412,	drum
7.673,	drum
7.890,	drum
8.984,	drum
9.292,	drum
6.908,	table lamp
7.053,	lighter
10.958,	torch
10.732,	cauldron
11.040,	goblet
11.664,	goblet
10.782,	cauldron
10.205,	cauldron
11.286,	coffee mug
12.546,	coffee mug
12.068,	torch
11.812,	torch
11.403,	torch
11.318,	goblet
11.198,	torch
10.351,	torch
10.463,	goblet
11.477,	goblet
10.985,	candle
8.176,	lampshade
5.905,	front curtain
6.226,	front curtain
6.438,	front curtain
6.348,	front curtain
6.194,	front curtain
6.574,	front curtain
6.461,	front curtain
7.079,	front curtain
6.876,	front curtain
7.276,	front curtain
6.827,	front curtain
7.621,	apron
6.337,	front curtain
6.850,	front curtain
6.438,	front curtain
6.500,	apron
5.826,	front curtain
5.363,	match
5.539,	match
5.619,	spotlight
5.494,	spotlight
5.497,	spotlight
5.744,	spotlight
5.598,	match
5.618,	spotlight
5.537,	spotlight
5.444,	spotlight
5.616,	spotlight
7.368,	front curtain
7.883,	front curtain
5.421,	spotlight
5.383,	front curtain
6.779,	front curtain
7.490,	front curtain
5.595,	spotlight
5.135,	match
8.931,	vacuum cleaner
9.416,	vacuum cleaner
10.049,	vacuum cleaner
9.830,	vacuum cleaner
9.725,	vacuum cleaner
9.250,	vacuum cleaner
9.561,	vacuum cleaner
9.738,	vacuum cleaner
9.309,	vacuum cleaner
8.841,	vacuum cleaner
9.707,	bath towel
10.451,	bath towel
9.035,	vacuum cleaner
9.322,	vacuum cleaner
9.674,	vacuum cleaner
10.060,	bath towel
8.765,	vacuum cleaner
6.887,	computer mouse
5.292,	quilt
4.914,	candle
5.577,	match
6.787,	match
6.161,	quilt
5.444,	quilt
6.033,	mosquito net
5.985,	mosquito net
6.933,	quilt
6.407,	quilt
7.831,	quilt
7.820,	quilt
6.470,	quilt
6.638,	quilt
7.042,	quilt
7.427,	quilt
5.869,	quilt
4.396,	lighter
6.722,	lampshade
6.678,	quilt
6.462,	quilt
6.866,	front curtain
6.537,	lampshade
6.498,	lampshade
7.459,	quilt
7.834,	quilt
6.892,	front curtain
6.692,	front curtain
7.218,	quilt
7.242,	quilt
6.859,	lampshade
6.892,	lampshade
7.240,	quilt
6.824,	bath towel
6.974,	lampshade
5.703,	spotlight
5.937,	mosquito net
6.642,	wig
13.067,	stage
12.018,	stage
7.328,	mosquito net
7.203,	mosquito net
5.418,	mosquito net
8.055,	tiger shark
7.007,	mosquito net
7.200,	mosquito net
6.148,	mosquito net
6.161,	paintbrush
7.425,	mosquito net
6.592,	toilet paper
6.937,	paintbrush
7.845,	paintbrush
6.767,	mosquito net
5.924,	spotlight
8.766,	water bottle
8.677,	pill bottle
9.612,	pill bottle
11.116,	spotlight
8.825,	water bottle
9.750,	water bottle
9.599,	water bottle
9.486,	water bottle
8.148,	spotlight
8.134,	spotlight
7.956,	spotlight
7.598,	spotlight
7.728,	water bottle
8.907,	water bottle
10.352,	water bottle
11.517,	water bottle
7.922,	spotlight
7.886,	spotlight
8.687,	cleaver
8.812,	cleaver
11.264,	stage
12.265,	stage
9.319,	cleaver
9.963,	cleaver
10.108,	cleaver
9.218,	spatula
9.169,	cleaver
9.209,	cleaver
10.191,	cleaver
11.338,	cleaver
9.701,	cleaver
9.644,	spatula
8.974,	cleaver
9.441,	cleaver
8.678,	cleaver
6.429,	pill bottle
8.492,	pill bottle
8.032,	candle
7.952,	stage
9.088,	stage
7.224,	candle
7.217,	pill bottle
7.810,	hockey puck
7.977,	hockey puck
8.021,	pill bottle
8.704,	pill bottle
8.811,	pill bottle
9.035,	pill bottle
9.422,	pill bottle
8.760,	pill bottle
8.645,	pill bottle
8.405,	pill bottle
8.337,	pill bottle
7.169,	pill bottle
7.739,	front curtain
10.099,	front curtain
11.041,	front curtain
11.088,	front curtain
7.287,	lampshade
6.705,	lampshade
8.272,	Granny Smith
8.122,	eraser
7.747,	lampshade
7.812,	lampshade
7.504,	front curtain
8.572,	front curtain
7.898,	lampshade
7.703,	lampshade
7.728,	spatula
7.923,	switch
8.246,	lampshade
7.384,	front curtain
6.582,	T-shirt
6.907,	velvet
10.928,	spotlight
11.858,	spotlight
6.597,	mosquito net
7.447,	mosquito net
6.833,	envelope
4.298,	shipwreck
7.479,	shower curtain
6.405,	shower curtain
7.312,	T-shirt
7.084,	shower curtain
7.607,	shower curtain
6.174,	shower curtain
5.254,	rapeseed
6.363,	wine bottle
6.524,	shower curtain
5.419,	spotlight
5.329,	shower curtain
7.687,	velvet
6.791,	velvet
9.860,	spotlight
7.237,	shower curtain
7.591,	shower curtain
6.570,	window shade
6.338,	window shade
6.647,	sink
7.066,	sink
5.958,	sink
6.143,	shower curtain
6.536,	sink
7.169,	sink
7.536,	toilet paper
7.417,	toilet paper
6.091,	sink
6.175,	spotlight
9.165,	fountain pen
7.601,	fountain pen
8.519,	spotlight
10.472,	spotlight
9.050,	fountain pen
10.077,	fountain pen
9.721,	fountain pen
7.925,	fountain pen
9.792,	pill bottle
9.579,	pill bottle
9.002,	pill bottle
8.307,	pill bottle
8.707,	fountain pen
8.613,	fountain pen
8.422,	toilet paper
9.889,	toilet paper
8.626,	fountain pen
8.124,	pill bottle
6.868,	wardrobe
7.461,	wardrobe
7.941,	spotlight
10.497,	spotlight
6.199,	wardrobe
6.926,	carton
8.270,	carton
7.633,	notebook computer
6.746,	wardrobe
6.816,	dishwasher
6.309,	wardrobe
6.410,	couch
6.961,	dishwasher
6.487,	dishwasher
7.415,	carton
8.878,	carton
7.284,	dishwasher
5.494,	spotlight
7.254,	pill bottle
7.161,	pill bottle
8.989,	stage
9.494,	stage
6.716,	carton
7.969,	carton
7.400,	wallet
8.125,	wallet
7.536,	lampshade
7.345,	table lamp
7.823,	pill bottle
7.040,	pill bottle
7.508,	lampshade
7.734,	chest
8.351,	chest
8.641,	chest
7.854,	lampshade
6.120,	lampshade
7.849,	water bottle
9.064,	water bottle
10.617,	water bottle
11.551,	water bottle
7.617,	hair spray
8.928,	water bottle
9.233,	water bottle
9.880,	water bottle
8.258,	hair spray
8.118,	hair spray
7.827,	hair spray
7.908,	hair spray
8.259,	hair spray
8.881,	hair spray
9.421,	water bottle
9.406,	hair spray
7.910,	hair spray
6.032,	lighter
5.574,	match
5.404,	match
5.269,	match
5.226,	spotlight
5.572,	match
5.543,	spotlight
5.523,	spotlight
5.299,	spotlight
4.956,	spotlight
4.573,	lighter
5.490,	front curtain
6.392,	front curtain
4.923,	spotlight
4.594,	spotlight
5.085,	front curtain
6.839,	front curtain
5.231,	match
4.965,	match
9.127,	banana
9.101,	jack-o'-lantern
8.137,	jack-o'-lantern
7.982,	jack-o'-lantern
11.372,	banana
9.682,	banana
8.877,	banana
8.916,	banana
11.321,	banana
10.062,	banana
9.167,	jack-o'-lantern
9.240,	jack-o'-lantern
13.585,	banana
13.163,	banana
12.669,	banana
11.153,	banana
11.025,	banana
12.131,	banana
8.141,	vacuum cleaner
7.982,	computer mouse
9.297,	computer mouse
9.786,	vacuum cleaner
7.655,	vacuum cleaner
8.465,	vacuum cleaner
9.062,	corkscrew
9.184,	corkscrew
7.218,	stethoscope
7.657,	stethoscope
7.697,	vacuum cleaner
8.658,	vacuum cleaner
7.158,	stethoscope
6.817,	vacuum cleaner
8.221,	corkscrew
9.260,	corkscrew
7.125,	vacuum cleaner
7.433,	lens cap
5.952,	lighter
7.818,	lighter
7.836,	water bottle
7.720,	syringe
5.568,	syringe
6.087,	water bottle
7.632,	water bottle
8.060,	water bottle
5.468,	acoustic guitar
5.749,	lampshade
5.674,	lampshade
6.197,	front curtain
6.009,	syringe
5.992,	syringe
7.455,	syringe
7.119,	syringe
5.958,	acoustic guitar
5.120,	Band-Aid

Many of the classes are completely unrelated to the objects found in the scenes. A very few classifications might be relevant, such as “banana”, “coffee cup”, “water bottle”, “orange”, and “bell pepper”. Interestingly, “bath towel” isn’t a class that is identified, although I would have expected it. For reference, see a list of the 1000 ImageNet classes.

Discussion

Based on how the classifications seem unrelated to the objects present in the scene, I think it was a mistake to assume that the ImageNet 1000 classes would cover most of the objects in the dataset’s scenes. The fact that the scenes don’t really map to ImageNet classes means that the experiment is a bit useless.

Next step

I’ll follow up in the minor increment experiment 2.0.1 by choosing crops such that the resulting images do contain objects covered by the ImageNet class list.