A Novel Method for Object Detection using Deep Learning and CAD

Models

Igor Garcia Ballhausen Sampaio

, Luigy Machaca

, Jos

e Viterbo

1 a

and Joris Gu

erin

2 b

Computing Institute, Universidade Federal Fluminense, Brazil

LAAS-CNRS, ONERA, Universit

e de Toulouse, France

Keywords:

Object Detection, CAD Models, Synthetic Image Generation, Deep Learning, Convolutional Neural Network.

Abstract:

Object Detection (OD) is an important computer vision problem for industry, which can be used for quality

control in the production lines, among other applications. Recently, Deep Learning (DL) methods have enabled

practitioners to train OD models performing well on complex real world images. However, the adoption of

these models in industry is still limited by the difﬁculty and the signiﬁcant cost of collecting high quality

training datasets. On the other hand, when applying OD to the context of production lines, CAD models of

the objects to be detected are often available. In this paper, we introduce a fully automated method that uses a

CAD model of an object and returns a fully trained OD model for detecting this object. To do this, we created

a Blender script that generates realistic labeled datasets of images containing the object, which are then used

for training the OD model. The method is validated experimentally on two practical examples, showing that

this approach can generate OD models performing well on real images, while being trained only on synthetic

images. The proposed method has potential to facilitate the adoption of object detection models in industry as

it is easy to adapt for new objects and highly ﬂexible. Hence, it can result in signiﬁcant costs reduction, gains

in productivity and improved products quality.

1 INTRODUCTION

Recently, Deep Learning (DL) has produced excellent

results for Object Detection (OD) (Liu et al., 2020).

On the one hand, a typical limitation with DL is the

requirement of large labeled datasets for training. In-

deed, although there are various large databases avail-

able online for OD, for speciﬁc industrial applications

it is always necessary to create custom datasets con-

taining the objects of interest. While the scenarios

present in public datasets are useful from both a re-

search and application standpoint, it was found that

industrial applications, such as bin picking or defect

inspection, have quite different characteristics that are

not modeled by the existing datasets (Drost et al.,

2017). As a result, methods that perform well on ex-

isting datasets sometimes show different results when

applied to industrial scenarios without retraining. The

process of generating a speciﬁc dataset for retraining

is tedious, and can be error-prone when conducted by

non-professional technicians. Moreover, generating a

https://orcid.org/0000-0002-0339-6624

https://orcid.org/0000-0002-8048-8960

new dataset and labeling it manually can be very time

consuming and expansive (Jabbar et al., 2017).

On the other hand, OD for industrial production

lines presents the speciﬁcity that the manufacturers

often have access to the CAD models of the objects

to detect. Thanks to advances in computer graphics

techniques, such as ray tracing (Shirley and Morley,

2003), the generation of photo-realistic images is now

possible. In such artiﬁcially generated images, the

computer can be employed to obtain bounding box la-

beling for free. The use of synthetic images rendered

from CAD models to train OD models has already

been proposed in (Peng et al., 2015), (Rajpura et al.,

2017) and (Hinterstoisser et al., 2018). However, their

approaches are not automated as they require manual

scene creation by Blender artists. In addition, the ob-

jects used in these works are usually generic, such as

buses, airplanes, cars or animals.

The main contribution of this paper is to present a

new method for training OD models in synthetic im-

ages generated from CAD models that is fully auto-

matic and thus well suited for industrial use. The pro-

posed method consists of the automatic generation of

realistic labeled images containing the objects to be

Sampaio, I., Machaca, L., Viterbo, J. and Guérin, J.

A Novel Method for Object Detection using Deep Learning and CAD Models.

DOI: 10.5220/0010451100750082

In Proceedings of the 23rd International Conference on Enterprise Information Systems (ICEIS 2021) - Volume 1, pages 75-82

ISBN: 978-989-758-509-8

detected, followed by the ﬁne-tuning of a pretrained

OD model on the artiﬁcial dataset. An extensive study

is conducted to properly select the user-deﬁned pa-

rameters so that it maximizes the performance on real

world images. Our method is evaluated using the

CAD models of two industrial objects for training, as

well as real labeled images containing the objects for

evaluation. The results obtained are very promising

as we manage to get F1-scores above 90% on real im-

ages while training only on synthetic images.

This paper is organized as follows. Section 2

presents the related work in the ﬁeld of object de-

tection and deep learning for industry. Section 3

provides detailed explanations about the proposed

method created. Section 4 describes our experiments,

presents the results obtained and discusses them. Fi-

nally, conclusions and directions for future work are

presented in Section 5.

2 RELATED WORK

This section presents related work about OD, indus-

trial applications of DL-based computer vision, as

well as computer vision methods using CAD models.

2.1 Object Detection

OD is a challenging computer vision problem that

consists in locating instances of objects from prede-

ﬁned categories in natural images (Prasad, 2012). It

has many applications in various domains such as

autonomous driving, security and medical diagno-

sis (Xiao et al., 2020). Deep learning techniques have

emerged as a powerful strategy for learning character-

istic representations directly from data and have led to

signiﬁcant advances in the ﬁeld of generic object de-

tection (Liu et al., 2020). In the last decade, many

competitions for object detection have been held to

provide large annotated datasets to the community,

and to unify the benchmarks and metrics for fair

comparison between proposed methods (Everingham

et al., 2010), (Lin et al., 2014), (Zhou et al., 2017),

(Kuznetsova et al., 2018).

Some examples of OD methods proposed within

the last few years include (He et al., 2015), where

the author proposes a new network structure, called

SPP-net, which can generate a ﬁxed-length represen-

tation, regardless of the size/scale of the image. Other

works such as (Jana et al., 2018) aim to improve pro-

cessing speed and at the same time efﬁciently identify

objects in the image. Finally, deeper CNNs have led

to record-breaking improvements in the detection of

more general object categories, a shift which came

about when DCNNs began to be successfully applied

to image classiﬁcation (Liu et al., 2020).

2.2 OD for Industrial Applications

Although general purpose OD methods have greatly

improved thanks to the availability of large public

datasets, the detection of instances in the industrial

context must be approached differently, since anno-

tated images are generally not available or rare. In-

deed, to train a deep learning model, hundreds of an-

notated images for each object category are needed.

Speciﬁc datasets need to be collected and annotated

for different target applications. This process is time-

consuming and laborious, and increases the burden on

operators, which goes against the goal of industrial

automation (Cohen et al., 2020), (Ge et al., 2020).

A public datasets adapted to the industrial context

was developed in (Drost et al., 2017). Unlike other

3D object detection datasets, this work models in-

dustrial waste collection and object inspection tasks

that often face different challenges. In addition, the

evaluation criteria are focused on practical aspects,

such as execution times, memory consumption, use-

ful measures of correction and precision. Other ex-

amples of datasets adapted to the industrial context in-

clude (Gu

erin et al., 2018b) and (Gu

erin et al., 2018a).

Finally, in (Yang et al., 2019), a method to detect

defects of tiny parts in real time was developed, based

on object detection and deep learning. To improve

their results, the authors consider the speciﬁcities of

the industrial application in their method such as the

properties of the parts, the environmental parameters

and the speed of movement of the conveyor. This is

a good example to adapt OD training methods to the

speciﬁc constraints of the industrial context.

2.3 CAD Models and OD

The ﬁrst commercial CAD programs came up in the

1970s, providing functions for 2D-drawing and data

archival, and evolved into the main engineering de-

sign tool (Lindsay et al., 2018), (Hirz et al., 2017).

These models can provide a scalable solution for

intelligent and automatic object recognition, track-

ing and augmentation based on generic object mod-

els (Ben-Himane et al., 2010). For example, CAD

models have been used to support multi-view detec-

tion (Zhang et al., 2013). In (Peng et al., 2015),

3D models were used as the primary source of infor-

mation to build object models. In other works, 3D

CAD models were used as the only source of labeled

data (Lin et al., 2014), (Everingham et al., 2010), but

they are limited to generic categories, such as cars and

ICEIS 2021 - 23rd International Conference on Enterprise Information Systems

motorcycles .

3 PROPOSED METHOD

An overview of this proposed method can be seen in

Figure 1. First, a custom Blender code is used to gen-

erate labeled training images containing the rendered

CAD model in context. Then, a pretrained object de-

tection model is ﬁne-tuned on the generated dataset.

Finally, the model can be used for inference on real

images (Figure 1b).

3.1 Image Generation

For the automatic generation of the training images,

the software Blender (Blender Online Community,

2018) is used. Blender is a powerful software for 3D

design, which includes features such as modeling, rig-

ging, simulation and rendering. Blender has a good

Python API, is open-source and has good GPU sup-

port.

In order to generate a synthetic training image

sample, our code requires several elements. First, a

CAD model of the object of interest as well as several

other industrial CAD models need to be available. In

the experiments of this paper, we use the two objects

shown in Figure 2, for which we also have real world

test images. The other objects serve as distractors

to help the model focusing on the right object. The

CAD models for the distractors are gathered from the

Grabcad website

. Different textures for the different

distractors as well as for the background are gathered

from the Poliigon website

. Finally, the color and tex-

ture of the object of interest are reproduced manually.

Once we have access to all the elements above,

the generation code goes as follows. A ﬂoor and a

table are created and some distractors are sampled.

Using physics simulation, the distractors are dropped

from a random height on the table. The position of

the object of interest is also randomly sampled. Once

the 3D scene is created, textures and colors are sam-

pled for the backgrounds and the distractors and the

entire scene is textured. Light sources and cameras

are also sampled and placed randomly. Constraints

on the camera pose are applied, in order to ensure that

the object appears in the camera view. Once the scene

has been created, the rendering occurs and generates

an image. By removing the light sources and mak-

ing the object of interest a light source itself, we can

generate another image which can be used for bound-

ing box labeling. This procedure is necessary because

https://grabcad.com/

https://www.poliigon.com/

even if we know the location of the object, it can be

partly hidden by distractors and thus distort the la-

beling. Example images generated using our Blender

code can be seen in Figure 3.

3.2 Model Training

In this work, we did not train a new CNN architecture

from the scratch. Instead, we used one of the pre-

trained models provided by TensorFlow Object De-

tection API (Huang et al., 2017). This approach is

called transfer learning and consists in starting train-

ing from a model that already knows basic feature ex-

traction skills and is less likely to overﬁt the synthetic

datasets. Indeed, the diversity that we can create with

Blender is limited as we cannot get an inﬁnite amount

of textures and distractors, and the diversity already

encountered by the network during pre-training can

help reduce overﬁtting. In addition, using a network

pre-trained on real images can prevent the network

from learning detection features that depend too much

on the generation procedure.

There exists several models in the TensorFlow

OD model zoo. More information on the perfor-

mance of the detection, as well as the reference ex-

ecution times, for each of the available pre-trained

models, can be found on the Github page of the

API

. In practice, the model used in this paper is the

faster rcnn inception v2 coco model, which provides

a good trade-off between performance and speed.

Faster R-CNN, the model used in this work, takes

as input an entire image and a set of object propos-

als. The network ﬁrst processes the whole image with

several convolutional and max pooling layers to pro-

duce a convolutional feature map. Then, for each

object proposal, a region of interest (RoI) pooling

layer extracts a ﬁxed-length feature vector from the

feature map. Each feature vector is fed into a se-

quence of fully connected layers that ﬁnally branch

into two sibling output layers: one that produces soft-

max probability estimates over K object classes plus

a catch-all “background” class and another layer that

outputs four real-valued numbers for each of the K

object classes. Each set of 4 values encodes reﬁned

bounding-box positions for one of the K classes. For

a more detailed view about Faster-RCNN, we refer

the reader to the original paper (Ren et al., 2015), or

to the following tutorial (Ananth, 2019).

To train the ﬁnal OD model, the TensorFlow OD

API requires a speciﬁc ﬁle structure of the training

images and labels. This step is carried out automati-

cally by our script.

https://github.com/tensorﬂow/models

A Novel Method for Object Detection using Deep Learning and CAD Models

(a) Training

(b) Inference

Figure 1: Overview of the proposed method for training an object detection network using a CAD model.

(a) Adblue (b) Yamaha logo

Figure 2: CAD models used in our experiments.

3.3 Parameter Selection Procedure

The Blender script used for image generation has

many hyperparameters that must be chosen before us-

ing it, such as the number of distractors, the number

of scenes generated or the resolution of synthetic im-

ages. Hence, we conduct a set of experiments to prop-

erly select these parameters in order to optimize the

OD results for inference on real images. In this sec-

tion, we explain the parameter selection procedure.

In other words, we present the dataset on which the

different sets of parameters were evaluated, as well as

the metrics used to assess the quality of the results ob-

tained with a given set of parameters. The results ob-

tained for this parameter selection procedure are pre-

sented in Section 4.

3.3.1 Test Dataset

The objective of this work is to validate that an object

detector trained on synthetic images can generalize

to real world industrial cases. Hence, we use a test

dataset composed of 380 real images containing the

objects corresponding to the CAD models used for

training. The bounding box annotation ﬁles for the

test images are generated manually using a software

called LabelImg

. This application allows us to draw

and save the annotations of each image as xml ﬁles in

the PASCAL VOC format (Everingham et al., 2010).

3.3.2 OD Metrics

In order to evaluate the quality of the trained model on

real images, and thus to be able to select the best hy-

perparameters for image generation and training, we

used standard OD metrics that are presented here.

Intersection over Union (IoU): is an evaluation

metric used to measure how much a predicted bound-

ing box matches with a ground truth bounding box.

For a pair of bounding boxes, IoU is deﬁned as the

area of the intersection divided by the area of the

union (Figure 4). If A corresponds to the ground-truth

box and B to the predicted box, then, IoU is computed

as :

IoU =

|A ∩ B|

|A ∪ B|

, (1)

where |.| denotes the area of a given shape. The nu-

merator is called the overlap area and the denomina-

tor is called the combined area. IoU ranges between 0

and 1, where 1 means that the bounding boxes are the

same and 0 that there is no overlap.

Precision, Recall, F1-Measure. We call conﬁ-

dence score, the probability that an anchor box con-

https://github.com/tzutalin/labelImg

ICEIS 2021 - 23rd International Conference on Enterprise Information Systems

(a) Adblue (b) Yamaha logo

Figure 3: Example of images generated using our custom Blender script.

Figure 4: Intersection over Union (IoU) computation.

tains an object from a certain class. It is usually pre-

dicted by the classiﬁer part of the object detector. The

conﬁdence score and IoU are used as the criteria to

determine whether a detection is a true positive or a

false positive. Given a minimal threshold on the con-

ﬁdence score for bounding box acceptance, and an-

other threshold on IoU to identify matching boxes, a

detection is considered a true positive (TP) if there

exists a ground truth such that: conﬁdence score >

threshold; the predicted class matches the class of the

ground truth; and IoU > threshold

IoU

. The violation

of any of the last two conditions generates a false pos-

itive (FP). In case multiple predictions correspond to

the same ground-truth, only the one with the highest

conﬁdence score counts as a true positive, while the

others are considered false positives. When a ground

truth bounding box is left without any matching pre-

dicted detection, it counts as a false negative (FN).

If we note TP, FP and TN respectively the number

of True Positives, False Positives and False Negatives

in a dataset, we can deﬁne the following metrics:

Precision =

T P

T P + FP

, (2)

Recall =

T P

T P + FN

, (3)

Precision · Recall

Precision + Recall

. (4)

A high precision means that most of the predicted

boxes had a corresponding ground truth, i.e., the ob-

ject detector is not producing bad predictions. A high

recall means that most of the ground truth boxes had

a corresponding prediction, i.e., the object detector

ﬁnds most objects in the images. The F

-Score is

the harmonic mean of the precision and recall, it is

needed when a balance between precision and recall

is sought.

In the case of object detection on production lines,

a low precision means that sometimes a part might be

absent and the model would not see it, whereas a low

recall means that sometimes the part is present and the

model raises an alert anyways. For this reason, both a

good recall and a precision are required and the choice

of using the F

-Score metric seems appropriate.

Average Precision. After an OD model has been

trained, the computation of Precision, Recall and F1-

score depends on the value of the two thresholds de-

ﬁned above (for the conﬁdence score and IoU). In

order to properly choose the values of these thresh-

olds, it is interesting to analyze the Precision x Re-

call curves. For each class, and for a given value of

the IoU threshold, the conﬁdence threshold is set as a

variable and sampled between 0 and 1 to plot a para-

metric curve with precision and recall as the x and

y-axis.

A class-speciﬁc object detector is considered good

if the precision remains high as the recall increases,

meaning that if you vary the conﬁdence limit, the pre-

cision and recall will still be high. Hence, to com-

pare between curves we generally rely on a numerical

metric called Average Precision (AP). Since 2010, the

A Novel Method for Object Detection using Deep Learning and CAD Models

standard computation method for AP consists in cal-

culating the area under the curve (AUC) of the Preci-

sion x Recall curve (Everingham et al., 2010).

4 EXPERIMENTS AND RESULTS

The results obtained for the parameter selection pro-

cedure as well as our ﬁnal evaluations are presented

here. These experiments were conducted on a Nvidia

Quadro P5000 GPU and a 2.90GHz Intel Xeon E3-

154M v5 processor (16 GB of RAM).

4.1 Hyperparameter Tuning

The parameter selection procedure is conducted ex-

clusively on the Yamaha logo object, the best set of

parameters is then tested on the Adblue object to en-

sure that it also performs well. The inﬂuence of four

tunable parameters on the ﬁnal results is studied here.

for each parameter, three values were selected for the

tests. These parameters and their studied values are:

• Resolution: 640x480, 960x540, 1080x720

• Camera poses: 2, 5, 20

• Number of scenes: 20, 50, 200

• Number of distractors: 0, 5, 20

From simple preliminary experiments that are not pre-

sented here, we concluded that the number of tex-

tures used for the ﬂoor, the distractors and the sup-

port should be set to the maximum number of textures

available (in our case 7 for the ﬂoor and 6 for the two

others). The parameter values used in this work were

chosen empirically, that is, after several test scenar-

ios, these values were the ones that generated the best

performance regarding the metrics.

In total, from the values selected for the four pa-

rameters, we sampled more than 30 combinations and

compared the OD results on the testing set of real im-

ages. For each combination tested, we trained the

Faster-RCNN CNN on the synthetic images that were

generated. We note that, for each hyperparameters

combination, the experiments were repeated 10 times

in order to attenuate the inﬂuence of the random com-

ponents in the generation and training process. For

reasons of space in this article, it was not possible to

present all results. However, in order to demonstrate

the importance of this parameter selection step, Ta-

ble 1 shows the best and the worst conﬁguration that

were tested.

In Table 1, we can see that the distractors are an

essential element in our proposed pipeline for image

generation. Indeed, when removing them, we can see

Table 1: Best and worst hyperparameters conﬁgurations ob-

tained and their corresponding results.

Parameters Best Case Worst Case

Resolution 960x540 960x540

cam poses 5 5

n scenes 20 20

n images 100 100

n distractors 20 0

Generation Time 1257.30 749.92

n samples 10 10

Precision %: Avg (Std.Dev) 78.06 (15.41) 57.34 (16.27)

Recall %: Avg (Std.Dev) 96.223 (4.07) 90.71 (6.45)

F1-Score %: Avg (Std.Dev) 85.19 (10.57) 66.80 (11.49)

a drop of around 23% in F1-score, on average across

the 10 experiment samples. We also tried combina-

tions with few distractors, but the F1-score results

dropped signiﬁcantly. This makes sense as the real

images evaluated had several distractors as well.

Another important point is that the resolution of

the images generated should be greater than the infer-

ence images. In all of our tests, this scenario always

produced the best results. It also makes sense as it is

easier to learn from a more detailed/complex model

and then evaluate in a less detailed/complex scenario.

Finally, we also tried to increase the number of

generated training images to see if this would lead

to an increase in performance. Surprisingly, we ac-

knowledged that the performance dropped for the

case with 20 distractors, 20 camera poses and 50

scenes (1000 images). This might mean that when

presented too many synthetic images, the model starts

overﬁtting to the biases involved by our generation

process, and it also indicates that we do not need a

large number of images to train our model. In ad-

dition to this performance drop, generating ten times

more images also makes the proposed pipeline almost

25 times slower (31037.16 seconds).

4.2 Results

In this section, we evaluate the results of the best com-

bination of parameters (Best Case from Table 1) in

more details. These results are presented in Table 2,

they correspond to using a conﬁdence threshold of

0.9 and an IoU threshold of 0.5. From Table 2, we

can see that the best parameters identiﬁed using only

the Yamaha logo produce similar results when applied

to another object (Adblue). This suggests that the

proposed parameters for our method seem to be well

suited for different objects and thus could generalize

well to various industrial use cases without additional

parameter tuning.

ICEIS 2021 - 23rd International Conference on Enterprise Information Systems

Table 2: Results obtained with the best set of hyperparame-

ters.

Object Precision % Recall % F1-Score %

Adblue 85.11 80.00 81.93

Yamaha logo 78.06 96.22 85.19

4.3 Discussion

It is difﬁcult to compare our results with other works

in the literature. Indeed, as far as we know, the ap-

proach presented in this work is the ﬁrst proposal to

build a fully automated pipeline that takes as input

the CAD model of an object and outputs a trained

object detection model for this object without any

real image. For fair comparison we would need to

compare our work with other end-to-end systematic

approaches to build OD models from CAD models,

which is impossible as it does not exists. Else, we

hope that the results presented in this work can serve

as a good baseline for comparison of future works in

this research direction.

However, we give a rough comparison with other

relevant works to give an idea of how well our ap-

proach is performing. In (Mazzetto et al., 2019) the

detection of objects in an automobile production line

was implemented, using only real images of the ob-

jects. In this work, the estimated detection accuracy

was around 90 %, which is only about 5 to 10 % bet-

ter than the results obtained in our work using only

synthetic images. In (Jabbar et al., 2017), the au-

thors also train an OD model using synthetic images

generated with Blender and evaluate the results in

real images. However, this approach is not entirely

automated since the scenes are created manually by

Blender artists to ensure photo-realism. The object

used in this work for evaluation is a glass of wine

and the maximum AP obtained is 71.14 %. We can

see that our systematic approach seems to work better

than this approach, however, we cannot reproduce the

method on our objects, as we cannot create the scenes

manually in the same way that they would. The po-

tential better performance of our approach can be ex-

plained by the fact that the loss of photo-realism can

be compensated by the higher number of images in

our synthetic datasets. Indeed, with our fully auto-

mated approach, it is faster and requires no effort to

generate more data, unlike in (Jabbar et al., 2017).

5 CONCLUSIONS

This work presents a systematic approach to train ob-

ject detection models to address industrial scenarios,

using only a CAD model of the object of interest as

input. The method ﬁrst generates realistic synthetic

images using a custom Blender script, and then trains

a faster-RCNN OD model using the TensorFlow OD

API. To understand and optimize the different param-

eters in the proposed pipeline, a systematic parame-

ter selection study is conducted using a Yamaha logo

CAD model for training and real images containing

the same object in context for evaluation. The se-

lected hyperparameters are then tested on an other

object, showing that they can generalize to different

scenarios.

Over the last decade, successful deep learning

methods have been developed to tackle the challeng-

ing problem of generic object detection. However,

when it comes to the problem of OD in an indus-

trial environment, the availability of good quality data

becomes a bottleneck. To address this issue we pro-

posed to use synthetic images for training, which is

challenging as it might not reﬂect the high variabil-

ity found in in real industrial environment (objects,

pieces and scenery, etc.). In addition, there is also a

difﬁculty in ﬁnding CAD models of speciﬁc indus-

trial objects so that they can be trained and other ap-

proaches can be tested and compared. Thus, as a con-

sequence of this work, a set of data was produced and

made publicly available for future research

Therefore, the main conclusion from this work is

that it is possible to train an object detection model on

a set of synthetic images generated from CAD mod-

els with excellent performance. In addition, it was

shown that a large set of images is not needed to ob-

tain a signiﬁcant result. Our experiments indicate that

the proposed rendering process is sufﬁcient to obtain

good performances and that the way of building and

rendering the scenes is crucial for the ﬁnal result.

ACKNOWLEDGEMENTS

Our work has beneﬁted from the AI Interdisciplinary

Institute ANITI. ANITI is funded by the French “In-

vesting for the Future – PIA3” program under the

Grant agreement n

◦

ANR-19-PI3A-0004.

REFERENCES

Ananth, S. (2019). Faster R-CNN for object detection, a

technical paper summary.

Ben-Himane, S., Hintestroisser, S., and Navab, N. (2010).

Computer vision CAD models. US Patent App.

12/682,199.

https://github.com/igorgbs/systematic approach cad

models

A Novel Method for Object Detection using Deep Learning and CAD Models

Blender Online Community (2018). Blender - a 3D mod-

elling and rendering package. Blender Foundation.

Cohen, J., Crispim-Junior, C., Grange-Faivre, C., and

Tougne, L. (2020). CAD-based learning for ego-

centric object detection in industrial context. In

15th International Conference on Computer Vision

Theory and Applications, volume 5, pages 644–651.

SCITEPRESS.

Drost, B., Ulrich, M., Bergmann, P., Hartinger, P., and Ste-

ger, C. (2017). Introducing MVTec ITODD-a dataset

for 3d object recognition in industry. In Proceedings

of the IEEE International Conference on Computer

Vision Workshops, pages 2200–2208.

Everingham, M., Van Gool, L., Williams, C. K., Winn,

J., and Zisserman, A. (2010). The pascal visual ob-

ject classes (VOC) challenge. International journal of

computer vision, 88(2):303–338.

Ge, C., Wang, J., Wang, J., Qi, Q., Sun, H., and Liao,

J. (2020). Towards automatic visual inspection: A

weakly supervised learning method for industrial ap-

plicable object detection. Computers in Industry,

121:103232.

erin, J., Gibaru, O., Nyiri, E., Thiery, S., and Palos,

J. (2018a). Automatic construction of real-world

datasets for 3D object localization using two cam-

eras. In IECON 2018-44th Annual Conference of

the IEEE Industrial Electronics Society, pages 3655–

3658. IEEE.

erin, J., Gibaru, O., Nyiri, E., Thieryl, S., and Boots,

B. (2018b). Semantically meaningful view selection.

In 2018 IEEE/RSJ International Conference on Intel-

ligent Robots and Systems (IROS), pages 1061–1066.

IEEE.

He, K., Zhang, X., Ren, S., and Sun, J. (2015). Spatial pyra-

mid pooling in deep convolutional networks for visual

recognition. IEEE transactions on pattern analysis

and machine intelligence, 37(9):1904–1916.

Hinterstoisser, S., Lepetit, V., Wohlhart, P., and Konolige,

K. (2018). On pre-trained image features and syn-

thetic images for deep learning. In Proceedings of the

European Conference on Computer Vision (ECCV).

Hirz, M., Rossbacher, P., and Gulanov

a, J. (2017). Future

trends in CAD–from the perspective of automotive

industry. Computer-aided design and applications,

14(6):734–741.

Huang, J., Rathod, V., Sun, C., Zhu, M., Korattikara, A.,

Fathi, A., Fischer, I., Wojna, Z., Song, Y., Guadar-

rama, S., et al. (2017). Speed/accuracy trade-offs for

modern convolutional object detectors. In Proceed-

ings of the IEEE conference on computer vision and

pattern recognition, pages 7310–7311.

Jabbar, A., Farrawell, L., Fountain, J., and Chalup, S. K.

(2017). Training deep neural networks for detecting

drinking glasses using synthetic images. In Interna-

tional Conference on Neural Information Processing,

pages 354–363. Springer.

Jana, A. P., Biswas, A., et al. (2018). YOLO based de-

tection and classiﬁcation of objects in video records.

In 2018 3rd IEEE International Conference on Recent

Trends in Electronics, Information & Communication

Technology (RTEICT), pages 2448–2452. IEEE.

Kuznetsova, A., Rom, H., Alldrin, N., Uijlings, J., Krasin,

I., Pont-Tuset, J., Kamali, S., Popov, S., Malloci, M.,

Duerig, T., et al. (2018). The open images dataset

v4: Uniﬁed image classiﬁcation, object detection, and

visual relationship detection at scale. arXiv preprint

arXiv:1811.00982.

Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ra-

manan, D., Doll

ar, P., and Zitnick, C. L. (2014). Mi-

crosoft COCO: Common objects in context. In Euro-

pean conference on computer vision, pages 740–755.

Springer.

Lindsay, A., Paterson, A., and Graham, I. (2018). Identi-

fying and quantifying inefﬁciencies within industrial

parametric CAD models. In Advances in Manufac-

turing Technology XXXII: Proceedings of the 16th In-

ternational Conference on Manufacturing Research,

volume 8, page 227. IOS Press.

Liu, L., Ouyang, W., Wang, X., Fieguth, P., Chen, J., Liu,

X., and Pietik

ainen, M. (2020). Deep learning for

generic object detection: A survey. International jour-

nal of computer vision, 128(2):261–318.

Mazzetto, M., Southier, L. F., Teixeira, M., and Casanova,

D. (2019). Automatic classiﬁcation of multiple ob-

jects in automotive assembly line. In 2019 24th

IEEE International Conference on Emerging Tech-

nologies and Factory Automation (ETFA), pages 363–

369. IEEE.

Peng, X., Sun, B., Ali, K., and Saenko, K. (2015). Learning

deep object detectors from 3D models. In Proceed-

ings of the IEEE International Conference on Com-

puter Vision, pages 1278–1286.

Prasad, D. K. (2012). Survey of the problem of object de-

tection in real images. International Journal of Image

Processing (IJIP), 6(6):441.

Rajpura, P. S., Bojinov, H., and Hegde, R. S. (2017). Ob-

ject detection using deep CNNs trained on synthetic

images. arXiv preprint arXiv:1706.06782.

Ren, S., He, K., Girshick, R., and Sun, J. (2015). Faster R-

CNN: Towards real-time object detection with region

proposal networks. In Advances in neural information

processing systems, pages 91–99.

Shirley, P. and Morley, R. K. (2003). Realistic ray tracing.

AK Peters/CRC Press.

Xiao, Y., Tian, Z., Yu, J., Zhang, Y., Liu, S., Du, S., and

Lan, X. (2020). A review of object detection based

on deep learning. Multimedia Tools and Applications,

pages 1–63.

Yang, J., Li, S., Wang, Z., and Yang, G. (2019). Real-time

tiny part defect detection system in manufacturing us-

ing deep learning. IEEE Access, 7:89278–89291.

Zhang, X., Yang, Y.-H., Han, Z., Wang, H., and Gao, C.

(2013). Object class detection: A survey. ACM Com-

puting Surveys (CSUR), 46(1):1–53.

Zhou, B., Lapedriza, A., Khosla, A., Oliva, A., and Tor-

ralba, A. (2017). Places: A 10 million image database

for scene recognition. IEEE transactions on pattern

analysis and machine intelligence, 40(6):1452–1464.

ICEIS 2021 - 23rd International Conference on Enterprise Information Systems