Partial half fine-tuning for object detection with unmanned aerial vehicles

IAES International Journal of Artificial Intelligence (IJ-AI)
Vol. 13, No. 1, March 2024, pp. 399~407
ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i1.pp399-407  399
Journal homepage: http://ijai.iaescore.com
Partial half fine-tuning for object detection with unmanned
aerial vehicles
Wahyu Pebrianto, Panca Mudjirahardjo, Sholeh Hadi Pramono
Department of Electrical Engineering, Faculty of Engineering, University of Brawijaya, Malang, Indonesia
Article Info ABSTRACT
Article history:
Received Oct 29, 2022
Revised Jan 16, 2023
Accepted Mar 10, 2023
Deep learning has shown outstanding performance in object detection tasks
with unmanned aerial vehicles (UAVs), which involve the fine-tuning
technique to improve performance by transferring features from pre-trained
models to specific tasks. However, despite the immense popularity of fine-
tuning, no works focused on to study of the precise fine-tuning effects of
object detection tasks with UAVs. In this research, we conduct an
experimental analysis of each existing fine-tuning strategy to answer which is
the best procedure for transferring features with fine-tuning techniques. We
also proposed a partial half fine-tuning strategy which we divided into two
techniques: first half fine-tuning (First half F-T) and final half fine-tuning
(Final half F-T). We use the VisDrone dataset for the training and validation
process. Here we show that the partial half fine-tuning: Final half F-T can
outperform other fine-tuning techniques and are also better than one of the
state-of-the-art methods by a difference of 19.7% from the best results of
previous studies.
Keywords:
Deep learning
Fine-tuning
Object detection
Unmanned aerial vehicles
VisDrone dataset
This is an open access article under the CC BY-SA license.
Corresponding Author:
Wahyu Pebrianto
Department of Electrical Engineering, Universitas Brawijaya
St. M.T. Haryono 167, Ketawanggede, Lowokwaru, Malang, East Java 65145
Email: wahyu.pebrianto1@gmail.com
1. INTRODUCTION
Object detection on intelligent machines integrated with drones and cameras or unmanned aerial
vehicles (UAVs) has been many applied to help various real-life domains, such as military [1], forest fire
detection [2], agriculture [3], security [4], [5], and urban surveillance [6]. That demands researchers in the field
of object detection to be able to analyze images obtained from UAVs. Currently, numerous research has
focused on deep learning [7] to overcome various object detection challenges, particularly in UAVs [8].
Deep learning advances are due to the availability of large-scale data, computing power such as
graphics processing unit (GPU), and continuous research. Since deep convolutional neural network (deep
CNN) has proposed by [9], deep learning methods have outperformed traditional machine learning in the
imageNet large scale visual recognition challenge (ILSVRC) 2010 challenge [10]. These results also encourage
many proposed other architecture such as visual geometry group (VGG) [11], GoogLeNet [12], and residual
networks (ResNets) [13], [14], cross stage partial network (CSPNet) [15], efficientNet [16] which are widely
used as backbone for feature extraction in classification and object detection tasks. In the object detection task,
region-based convolutional neural network (R-CNN) is the first deep learning-based object detection method
[17] that has outperformed other traditional detectors such as deformable parts model (DPM) [18] and SegDPM
[19] in the challenge PASCAL visual object classes (PASCAL VOC) 2010 [20]. Other popular methods, such
as the two-stage detector: Fast R-CNN [21], faster R-CNN [22], and the one-stage detector: you only look once
(YOLO) [23]–[28], retinaNet [29], single shot multibox detector (SSD) [30] which have demonstrated strong

 ISSN: 2252-8938
Int J Artif Intell, Vol. 13, No. 1, March 2024: 399-407
400
detection performance on the general object, such as Microsoft common objects in context (COCO) [31],
PASCAL VOC, and imageNet. But those methods average it has poor performance when detecting small
objects, which is the characteristic of the data captured by the UAVs. Therefore requires the right adjustment
strategy in the model architecture or the right training strategy for object detection tasks in the UAVs. Much
of the research recently has been proposed to overcome the challenge of object detection in UAVs. Such [32]
used a deformable convolutional layer [33] in the last three stages of ResNet50, which also adopted a cascade
architecture and augmentation data to improve model performance when detecting small and dense objects.
[34] proposed RRNet with adaptive resampling as an augmentation technique followed by a regression module
to improve bounding box prediction accuracy. The current work also involves transformer architecture which
many use in natural language processing tasks [35]. Such as [36] that replaced the YOLOv5 detection layer
with the detection of transformers, and [37] proposed a ViT-YOLO that combined multi-head self-attention
[38] in the original CSP-Darknet backbone YOLOv4-P7 [27], bidirectional feature pyramid network (BiFPN)
[39], and YOLOv3 as head detection layers. ViT-YOLO obtained an mean average precision (mAP) of 39.41%
and was one of the top results in the dataset VisDrone2021-Det 2021 challenge [40]. The results not be
separated by the availability of annotated data, which has become the key to the training process and
benchmark. However, problems arise if the amount of annotated data is insufficient for the training process. It
can easily be overfitting during the training process. Therefore also have an impact on the performance of the
object detection method. While in practical situations, availability annotated data generally is difficult to obtain.
The current popular solution is to use a fine-tuning technique [41]. Fine-tuning is a common technique
that takes advantage of the pre-trained model on large-scale data. Then transferred its features to new tasks
with fewer data. This technique has been evident to increase the generalization of the model and avoid
overfitting [42]–[47]. Such as has been done [17], [21]–[23], [29], [30] in general object detection tasks and
[34], [36], [37], [48] in object detection tasks with UAVs. However, despite the immense popularity of fine-
tuning techniques, no research focuses to studies the precise fine-tuning effects for object detection tasks with
UAVs. Several reasons: i) the current work uses fine-tuning techniques without focus observing the impact of
the transferred features or building an efficient fine-tuning strategy. ii) the differences in data sizes along with
features are crucial to consider in the fine-tuning process. For example, the COCO [31] dataset and VisDrone
[40] dataset have differences, as described in Figure 1(a) VisDrone which dominated by small objects, while
in Figure 1(b) COCO dataset is dominated by general objects. The differences in the characteristics of these
features can have impacted the performance of the fine-tuning technique, and very important to investigate
which is the best fine-tuning strategy in object detection tasks with UAVs.
Small Object Density Different Illumination
(a)
General Object
(b)
Figure 1. Image from dataset, (a) VisDrone dataset and (b) COCO dataset
In this study, we conduct experimental analysis on every existing fine-tuning approach, such as
common fine-tuning (Common F-T) and frozen fine-tuning (Frozen F-T). We also propose a partial half fine-
tuning strategy which consists of two techniques: first half fine-tuning (First Half F-T) and final half fine-
tuning (Final half F-T). The first half F-T freezes the first half layer in the target network backbone, and the
Final half F-T freezes the final half layer. The remaining layers are not frozen or randomly initialized during
the training process for the target-task. In the evaluation process, we compared the partial half fine-tuning
strategy with Frozen F-T, Common F-T, traditional training (Traditional-T), and one of the state-of-the-art
methods of object detection tasks with UAVs. The main contributions of this paper are,

Int J Artif Intell ISSN: 2252-8938 
Partial half fine-tuning for object detection with unmanned aerial vehicles (Wahyu Pebrianto)
401
– We propose a partial half fine-tuning strategy: First half F-T and Final half F-T as a fine-tuning strategy
for object detection with UAVs. (Section 2)
– We show the experimental results to answer where is the best procedure for transferring features
from the base-task to the target-task, whether Common F-T, Frozen F-T, or partial half fine-tuning.
(Sections 3.4.1-3.4.2)
– We show that partial half fine-tuning can outperform one of the state-of-the-art methods in object
detection tasks with UAVs. (Section 3.4.3).
2. RESEARCH METHOD
In this section, we set up several fine-tuning procedures along with a partial half fine-tuning strategy
to transfer a set of learned features in base-task and use them for target-task, as shown in Figure 2. That consists
of three concepts: base-task, target-task, and transfer or fine-tuning. In base-task is shown in Figure 2(a): given
the base-dataset 𝐵𝑑𝑎𝑡𝑎𝑠𝑒𝑡, which is trained with the base-network 𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 to complete the base-task 𝐵𝑡𝑎𝑠𝑘. In
target-task is shown in Figures 2(b)-2(e): given a target-dataset 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 to be trained with target-network
𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 for the target-task 𝑇𝑡𝑎𝑠𝑘. Then in the fine-tuning process, is do transfer features in the form of weight
parameter 𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 from the pre-trained model to improve 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 performance on new task 𝑇𝑡𝑎𝑠𝑘. Where
𝐵𝑑𝑎𝑡𝑎𝑠𝑒𝑡 ≠ 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 or 𝐵𝑡𝑎𝑠𝑘 ≠ 𝑇𝑡𝑎𝑠𝑘
Figure 2. Fine-tuning strategy, (a) The base model in base-task uses a pre-trained model, (b) Common F-T,
(c) Frozen F-T, (d) First half F-T, and (e) Final half F-T

 ISSN: 2252-8938
402
In 𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 we leverage pre-trained YOLOv5 [49] on 𝐵𝑑𝑎𝑡𝑎𝑠𝑒𝑡 COCO [31] in 𝐵𝑡𝑎𝑠𝑘. The total layers
in the backbone YOLOv5 are 10 blocks which we denote as 𝐵0−9. Then a fine-tuning process is carried out to
𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 by training it to 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 VisDrone [41] to complete the 𝑇𝑡𝑎𝑠𝑘 object detection on the UAVs. The
details of each fine-tuning strategy, including the spatial half fine-tuning: First half F-T, and Final half F-T that
we propose illustrated in Figure 2. With the following details,
– Common fine-tuning (Common F-T): transfer feature 𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 from pre-trained model to 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 with
condition parameter feature 𝐵0−9 on 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 initialized randomly when trained with 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 in specific
𝑇𝑡𝑎𝑠𝑘. The details of the illustration are explained in Figure 2(b).
– Frozen fine-tuning (Frozen F-T): transfers feature 𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 from pre-trained model to 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 with
condition 𝐵0−9 on 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 frozen, meaning that there is no process of changing parameters in 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘
during training process with 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 for 𝑇𝑡𝑎𝑠𝑘. The details of the illustration are explained in Figure 2(c).
– First half fine-tuning (First half F-T): transfers feature 𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 from pre-trained model to 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 with
condition 𝐵0−4 on 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 frozen. Then 𝐵5−9 initialized randomly, meaning that during the training
process on 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 on 𝑇𝑡𝑎𝑠𝑘 there was no process of changing the parameters of 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 on 𝐵0−4.
However, in 𝐵5−9 there are changes during the training process. The details of the illustration are
explained in Figure 2(d).
– Final half fine-tuning (Final half F-T): is the opposite of First half F-T, in Final half F-T transfer feature
𝐵𝑛𝑒𝑡𝑤𝑜𝑟𝑘 from pre-trained model to 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 with condition 𝐵5−9 on 𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 in frozen, and 𝐵0−4
initialized randomly, where during the training process with 𝑇𝑑𝑎𝑡𝑎𝑠𝑒𝑡 on 𝑇𝑡𝑎𝑠𝑘 parameter changes to
𝑇𝑛𝑒𝑡𝑤𝑜𝑟𝑘 only occur in 𝐵0−4. The details of the illustration are explained in Figure 2(e).
3. EXPERIMENTAL
3.1. Dataset
To validate the effectiveness of the fine-tuning strategy, we used the dataset VisDrone2019-Det [50].
The VisDrone2019-Det dataset is the same as VisDrone2021-Det [40]. The dataset consists of ten object
categories: (pedestrian, person, bicycle, car, van, truck, tricycle, awning-tricycle, bus, and motorcycle) with a
total number of 10,209 images, 6,471 for training, 548 for validation, and 3,190 for testing. For all the fine-
tuning techniques in this research, we train that on the VisDrone training set and evaluate them with a validation
set.
3.2. Evaluation metrics
For the evaluation process of each fine-tuning strategy, we used several relevant parameters [51].
Including precision (P), recall (R), average precision (AP), and mean average precision (mAP). We measured
mAP by setting the threshold value to 0.5 intersections over union (IoU). The details of each parameter are
explained in (1)-(4), (1) explains P, (2) explains R, (3) explains AP, and (4) explains mAP.
Precision (P) =
TP
TP+FP
(1)
Recall (R) =
TP
TP+FN
(2)
AP = ∑ (𝑅𝑛+1 − 𝑅𝑛)
𝑛 𝑃(𝑅
̃)
𝑅
̃:𝑅
̃≥𝑅𝑛+1
𝑚𝑎𝑚𝑎𝑥𝑠𝑎𝑥
(3)
mAP =
1
N
∑ 𝐴𝑃𝑖
N
i=1 (4)
Based on (1) and (2), where true positive (TP) is the detection correct from the ground truth bounding
box, false positive (FP) is the object that was detected but misplaced, and false negative (FN) is the ground
truth of the bounding box not identified. AP is the average value of P and R as shown in (3) 𝑃(𝑅
̃) is the
measured P at R. Then mAP is the average of AP used to measure all class categories in the dataset and is a
metric used to measure the accuracy of object detection. As shown in (4) 𝐴𝑃𝑖 is the AP in class 𝑖, and 𝑁 is the
total number of classes evaluated.
3.3. Experimental details
We use the pre-trained model YOLOv5x [49] as a base-task in the COCO dataset [31] same as that
used in [36]. We transfer features from pre-trained models to each fine-tuning technique: Common F-T, Frozen
F-T, First half F-T, and Final half F-T. Then we train each that technique to the VisDrone target-dataset [50]
as an object detection task in UAVs. In the training phase, we use stochastic gradient descent as optimization

403
with momentum 0.9, batch size 16, learning rate 0.01, and training iterations of 50 epochs with 640×640 input.
The framework for the training and validation process uses PyTorch with a Tesla T4 GPU.
3.4. Results and discussion
To reflect the scalability of each fine-tuning strategy. We analyze each fine-tuning approach with an
input scale of 640×640, and we also involve traditional training (Traditional-T), that shown in (section 3.4.1).
We analyze the training results with different input scales, shown in (section 3.4.2). In the last, we compare
the best results in this study with one of the state-of-the-art methods that used the VisDrone validation set as
the evaluation process and with the same input scale, shown in (section 3.4.3).
3.4.1. Analysis effect fine-tuning strategy
Table 1 presents the validation results for each fine-tuning strategy with an input scale of 640×640.
Common F-T can achieve detection mAP accuracy 8.9% greater than Traditional-T. The deferences result of
8.9% from Traditional-T means Common F-T indicated more increases in generalization of the model than
Frozen F-T that only increases by 3.5% mAP accuracy compared to Traditional-T. The low improvement from
Frozen F-T that because the features transferred from the COCO base-task directly adjusted to the VisDrone
target-task without setting random parameters, and no feature update during the training process to the target-
task. For partial half fine-tuning: Final half F-T can achieve detection mAP accuracy 9.3% greater than
Traditional-T, 5.8% from Frozen F-T, and at the same time also outperform Common F-T with a deference
0.4% more height, that results prove Final half F-T is the best strategy for fine-tuning in input scale of 640×640.
But, the mAP accuracy for the first half F-T result of 5% is slightly lower than the result of Common F-T and
5.4% lower than the Final half F-T. Base on research conducted by Yosinski et al. [41] that shown a transition
process when the features of the base-task transferred to the target-task. In our study, the low improvement
from first half F-T that because the first half layer in the target-network learns general features but is more
specific to the COCO dataset than to the VisDrone dataset in the target-task. While the Final half F-T proves
that the last half layer is more common or matches the features of the COCO base task and VisDrone target
task. The details of the detection from Table 1 described in Table 2. While Figure 3 shows one of the results
of the visualization of detection with a validation set.
Table 1. Evaluation results with VisDrone validation set
Training Input size Precision Recall mAP_0.5
Traditional-T 640×640 43.8 33.8 32.8
Common F-T 640×640 53.3 40.9 41.7
Frozen F-T 640×640 47 36.8 36.3
First half F-T 640×640 48.3 36.6 36.7
Final half F-T 640×640 52.4 41.4 42.1
Table 2. Detection results with VisDrone validation set
Model Pedestrian People Bicycle Car Van Truck Tricycle Awn Bus Motor
Traditional-T 42.2 33.3 09 73.3 34.9 27.7 17.3 10.1 41 39.4
Common F-T 49.9 39.1 20.2 78.6 43.1 40.2 28.8 15.7 55.6 46.1
Frozen F-T 42.8 35.3 12.3 74.6 38.3 34.1 23.5 12.7 49.4 39.5
First half F-T 42.7 35.5 11.6 75.3 36.8 37.7 23.2 12.7 50.1 41
Final half F-T 49.9 40.1 19.4 78.6 44 40.1 30.2 15.5 54.8 48.3
3.4.2. Analysis on different scale
To obtain a more in-depth analysis, we evaluated each fine-tuning technique with different input
scales, namely 416×416, 608×608, 832×832, and 960×960, as described in Table 3. Common F-T outperforms
the results of Traditional-T, Frozen F-T, First half F-T, and particularly, Final half F-T by a difference of 0.1%
on 416×416 and 608×608 scales. However, on the 832×832 and 960×960 scales, the final half F-T is superior
to Traditional-T, Frozen F-T, first half F-T, and especially Common F-T by a difference of 0.5% on both scales.
These results indicate that the Common F-T is slightly higher than the Final half F-T with a smaller input scale.
However, based on our experiment, the results of Final half F-T are more robust with higher input scales than
Common F-T, Traditional-T, Frozen F-T, and First half F-T.
3.4.3. Compare with state-of-the-art
In this section, we compare our proposed partial half fine-tuning: Final half F-T with one of the state-
of-the-art methods employed by [48]. We only compared it with one previous work because the authors used
the same input scale and VisDrone validation set for the evaluation process. We focus on the 832×832 scale

 ISSN: 2252-8938
404
input because it is the best result from [48]. Such as described in Table 4, the results of Final half F-T are 20.3%
greater than SlimYOLOv3-SPP3-50 and 19.7% than YOLOv3-SPP3. That result indicates our proposed Final
half F-T is better than one of the previous studies.
Figure 3. Some visualization results from our research
Table 3. Validation result at different scales using the VisDrone validation set
Model Input Precision Recall mAP 0.5
Traditional-T 416×416 36.6 25.5 24.2
608×608 42.5 32.6 31.9
832×832 44.1 38 36.3
960×960 45 38.1 37.1
416×416 46.4 31.7 31.9
Common F-T 608×608 52.7 40.4 41.1
832×832 56.6 44.3 45.6
960×960 57.7 45.6 46.8
416×416 39.4 26.6 25.9
Frozen F-T 608×608 45.6 35.7 35.3
832×832 49.3 40.9 40.4
960×960 49.8 43.2 42.5
416×416 40.8 26.9 26.7
First half F-T 608×608 49.1 35.2 35.9
832×832 51.9 41.3 41.4
960×960 53.1 43.5 43.4
416×416 44.6 31.5 31.8
Final half F-T 608×608 51.6 39.9 41
832×832 56.4 44.6 46.1
960×960 55.8 46.3 47.3

405
Table 4. Results of comparison with one of the state-of-the-art methods
Model Input Precision Recall mAP 0.5
SlimYOLOv3-SPP3-50 832×832 45.9 36 25.8
YOLOv3-SPP3 832×832 43.5 38 26.4
Final half F-T 832×832 56.4 44.6 46.1
4. CONCLUSION
In this study, we conduct experimental analysis on every existing fine-tuning approach and propose a
partial half fine-tuning strategy which consists of two techniques: First half F-T and Final half F-T. In the
evaluation process, we used the VisDrone validation set. Here we show that the result of Final half F-T can
achieve detection mAP accuracy 9.3% greater than Traditional-T, 5.8% from Frozen F-T, and 0.4% from
Common F-T in an input scale of 640×640, and its also more accurate at higher scales, such as in scale 832×832
and 960×960. Then we compared the final half F-T with one of the state-of-the-art methods, based on the mAP
IoU 0.5 and the same 832×832 input scale. Here we show that the results of final half F-T are 20.3% greater than
SlimYOLOv3-SPP3-50 and 19.7% than YOLOv3-SPP3. That means our technique is better than other fine-
tuning techniques and also better than one of the state-of-the-art methods in object detection with UAVs.
REFERENCES
[1] P. Gupta, B. Pareek, G. Singal, and D. V. Rao, “Edge device based Military Vehicle Detection and Classification from UAV,”
Multimedia Tools and Applications, vol. 81, no. 14, pp. 19813–19834, 2022, doi: 10.1007/s11042-021-11242-y.
[2] C. Yuan, Z. Liu, and Y. Zhang, “Aerial Images-Based Forest Fire Detection for Firefighting Using Optical Remote Sensing
Techniques and Unmanned Aerial Vehicles,” Journal of Intelligent & Robotic Systems, vol. 88, no. 2–4, pp. 635–654, Dec. 2017,
doi: 10.1007/s10846-016-0464-7.
[3] U. R. Mogili and B. B. V. L. Deepak, “Review on Application of Drone Systems in Precision Agriculture,” Procedia Computer
Science, vol. 133, pp. 502–509, 2018, doi: 10.1016/j.procs.2018.07.063.
[4] A. Singh, D. Patil, and S. N. Omkar, “Eye in the Sky: Real-Time Drone Surveillance System (DSS) for Violent Individuals
Identification Using ScatterNet Hybrid Deep Learning Network,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern
Recognition Workshops (CVPRW), Jun. 2018, vol. 2018-June, pp. 1710–17108, doi: 10.1109/CVPRW.2018.00214.
[5] X. Zhang and K. Kusrini, “Autonomous long-range drone detection system for critical infrastructure safety,” Multimedia Tools and
Applications, vol. 80, no. 15, pp. 23723–23743, Jun. 2021, doi: 10.1007/s11042-020-10231-x.
[6] W. L. Leong, N. Martinel, S. Huang, C. Micheloni, G. L. Foresti, and R. S. H. Teo, “An Intelligent Auto-Organizing Aerial Robotic
Sensor Network System for Urban Surveillance,” Journal of Intelligent & Robotic Systems, vol. 102, no. 2, p. 33, Jun. 2021, doi:
10.1007/s10846-021-01398-y.
[7] Y. Lecun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015, doi: 10.1038/nature14539.
[8] X. Wu, W. Li, D. Hong, R. Tao, and Q. Du, “Deep Learning for Unmanned Aerial Vehicle-Based Object Detection and Tracking:
A survey,” IEEE Geoscience and Remote Sensing Magazine, vol. 10, no. 1, pp. 91–124, Mar. 2022, doi:
10.1109/MGRS.2021.3115137.
[9] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” in Neural
Information Processing Systems, 2012, pp. 1106–1114.
[10] O. Russakovsky et al., “ImageNet Large Scale Visual Recognition Challenge,” International Journal of Computer Vision, vol. 115,
no. 3, pp. 211–252, 2015, doi: 10.1007/s11263-015-0816-y.
[11] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” 3rd International
Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings, 2015.
[12] C. Szegedy et al., “Going deeper with convolutions,” Proceedings of the IEEE Computer Society Conference on Computer Vision
and Pattern Recognition, vol. 07-12-June, pp. 1–9, 2015, doi: 10.1109/CVPR.2015.7298594.
[13] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE Computer Society
Conference on Computer Vision and Pattern Recognition, vol. 2016-Decem, pp. 770–778, 2016, doi: 10.1109/CVPR.2016.90.
[14] K. He, X. Zhang, S. Ren, and J. Sun, “Identity mappings in deep residual networks,” Lecture Notes in Computer Science (including
subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9908 LNCS, pp. 630–645, 2016, doi:
10.1007/978-3-319-46493-0_38.
[15] C.-Y. Wang, H.-Y. Mark Liao, Y.-H. Wu, P.-Y. Chen, J.-W. Hsieh, and I.-H. Yeh, “CSPNet: A New Backbone that can Enhance
Learning Capability of CNN,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW),
Jun. 2020, vol. 2020-June, pp. 1571–1580, doi: 10.1109/CVPRW50498.2020.00203.
[16] M. Tan and Q. V. Le, “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” May 2019, doi:
10.48550/arXiv.1905.11946.
[17] R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic
segmentation,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 580–587,
2014, doi: 10.1109/CVPR.2014.81.
[18] P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and D. Ramanan, “Object Detection with Discriminatively Trained Part-Based
Models,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 32, no. 9, pp. 1627–1645, Sep. 2010, doi:
10.1109/TPAMI.2009.167.
[19] S. Fidler, R. Mottaghi, A. Yuille, and R. Urtasun, “Bottom-Up Segmentation for Top-Down Detection,” in 2013 IEEE Conference
on Computer Vision and Pattern Recognition, Jun. 2013, pp. 3294–3301, doi: 10.1109/CVPR.2013.423.
[20] M. Everingham, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman, “The Pascal Visual Object Classes (VOC) Challenge,”
International Journal of Computer Vision, vol. 88, no. 2, pp. 303–338, Jun. 2010, doi: 10.1007/s11263-009-0275-4.
[21] R. Girshick, “Fast R-CNN,” Proceedings of the IEEE International Conference on Computer Vision, vol. 2015 Inter, pp. 1440–
1448, 2015, doi: 10.1109/ICCV.2015.169.

 ISSN: 2252-8938
406
[22] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,”
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137–1149, 2017, doi:
10.1109/TPAMI.2016.2577031.
[23] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” in 2016 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2016, pp. 779–788, doi: 10.1109/CVPR.2016.91.
[24] J. Redmon and A. Farhadi, “YOLO9000: Better, faster, stronger,” Proceedings - 30th IEEE Conference on Computer Vision and
Pattern Recognition, CVPR 2017, vol. 2017-Janua, pp. 6517–6525, 2017, doi: 10.1109/CVPR.2017.690.
[25] J. Redmon and A. Farhadi, “YOLOv3: An Incremental Improvement,” 2018, [Online]. Available: http://arxiv.org/abs/1804.02767.
[26] W. Pebrianto, P. Mudjirahardjo, and S. H. Pramono, “YOLO Method Analysis and Comparison for Real-Time Human Face
Detection,” in 2022 11th Electrical Power, Electronics, Communications, Controls and Informatics Seminar (EECCIS), Aug. 2022,
pp. 333–338, doi: 10.1109/EECCIS54468.2022.9902919.
[27] A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, “YOLOv4: Optimal speed and accuracy of object detection,” 2020, [Online].
Available: http://arxiv.org/abs/2004.10934.
[28] C. Y. Wang, A. Bochkovskiy, and H. Y. M. Liao, “Scaled-yolov4: Scaling cross stage partial network,” Proceedings of the IEEE
Computer Society Conference on Computer Vision and Pattern Recognition, pp. 13024–13033, 2021, doi:
10.1109/CVPR46437.2021.01283.
[29] T. Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollar, “Focal loss for dense object detection,” IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 42, no. 2, pp. 318–327, 2020, doi: 10.1109/TPAMI.2018.2858826.
[30] W. Liu et al., “SSD: Single Shot MultiBox Detector,” in Lecture Notes in Computer Science (including subseries Lecture Notes in
Artificial Intelligence and Lecture Notes in Bioinformatics), 2016, pp. 21–37.
[31] T. Y. Lin et al., “Microsoft COCO: Common objects in context,” Lecture Notes in Computer Science (including subseries Lecture
Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 8693 LNCS, no. PART 5, pp. 740–755, 2014, doi:
10.1007/978-3-319-10602-1_48.
[32] X. Zhang, E. Izquierdo, and K. Chandramouli, “Dense and small object detection in UAV vision based on cascade network,”
Proceedings - 2019 International Conference on Computer Vision Workshop, ICCVW 2019, pp. 118–126, 2019, doi:
10.1109/ICCVW.2019.00020.
[33] J. Dai et al., “Deformable Convolutional Networks,” in 2017 IEEE International Conference on Computer Vision (ICCV), Oct.
2017, vol. 2017-Octob, pp. 764–773, doi: 10.1109/ICCV.2017.89.
[34] C. Chen et al., “RRNet: A Hybrid Detector for Object Detection in Drone-Captured Images,” in 2019 IEEE/CVF International
Conference on Computer Vision Workshop (ICCVW), Oct. 2019, pp. 100–108, doi: 10.1109/ICCVW.2019.00018.
[35] S. Khan, M. Naseer, M. Hayat, S. W. Zamir, F. S. Khan, and M. Shah, “Transformers in Vision: A Survey,” ACM Computing
Surveys, vol. 54, no. 10s, pp. 1–41, Jan. 2022, doi: 10.1145/3505244.
[36] X. Zhu, S. Lyu, X. Wang, and Q. Zhao, “TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object
Detection on Drone-captured Scenarios,” in 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW),
Oct. 2021, pp. 2778–2788, doi: 10.1109/ICCVW54120.2021.00312.
[37] Z. Zhang, X. Lu, G. Cao, Y. Yang, L. Jiao, and F. Liu, “ViT-YOLO:Transformer-Based YOLO for Object Detection,” in 2021
IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), Oct. 2021, vol. 2021-Octob, pp. 2799–2808, doi:
10.1109/ICCVW54120.2021.00314.
[38] A. Vaswani et al., “Attention is all you need,” Advances in Neural Information Processing Systems, pp. 5999–6009, 2017.
[39] M. Tan, R. Pang, and Q. V. Le, “EfficientDet: Scalable and Efficient Object Detection,” in 2020 IEEE/CVF Conference on
Computer Vision and Pattern Recognition (CVPR), Jun. 2020, pp. 10778–10787, doi: 10.1109/CVPR42600.2020.01079.
[40] Y. Cao et al., “VisDrone-DET2021: The Vision Meets Drone Object detection Challenge Results,” in 2021 IEEE/CVF International
Conference on Computer Vision Workshops (ICCVW), Oct. 2021, vol. 2021-Octob, pp. 2847–2854, doi:
10.1109/ICCVW54120.2021.00319.
[41] J. Yosinski, J. Clune, Y. Bengio, and H. Lipson, “How transferable are features in deep neural networks?,” in Neural Information
Processing Systems, 2014, vol. 27, pp. 3320–3328.
[42] W. Ouyang, X. Wang, C. Zhang, and X. Yang, “Factors in Finetuning Deep Model for Object Detection with Long-Tail
Distribution,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2016, vol. 2016-Decem, pp.
864–873, doi: 10.1109/CVPR.2016.100.
[43] X. Li, Y. Grandvalet, and F. Davoine, “Explicit inductive bias for transfer learning with convolutional networks,” in 35th
International Conference on Machine Learning, ICML 2018, Feb. 2018, vol. 6, pp. 4408–4419.
[44] M. Wortsman et al., “Robust fine-tuning of zero-shot models,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern
Recognition (CVPR), Jun. 2022, pp. 7949–7961, doi: 10.1109/CVPR52688.2022.00780.
[45] S. X. Hu, D. Li, J. Stuhmer, M. Kim, and T. M. Hospedales, “Pushing the Limits of Simple Pipelines for Few-Shot Learning:
External Data and Fine-Tuning Make a Difference,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition
(CVPR), Jun. 2022, pp. 9058–9067, doi: 10.1109/CVPR52688.2022.00886.
[46] S. Niu, Y. Liu, J. Wang, and H. Song, “A Decade Survey of Transfer Learning (2010–2020),” IEEE Transactions on Artificial
Intelligence, vol. 1, no. 2, pp. 151–166, Oct. 2020, doi: 10.1109/TAI.2021.3054609.
[47] G. S. Dhillon, P. Chaudhari, A. Ravichandran, and S. Soatto, “A Baseline for Few-Shot Image Classification,” Sep. 2019.
[48] P. Zhang, Y. Zhong, and X. Li, “SlimYOLOv3: Narrower, Faster and Better for Real-Time UAV Applications,” in 2019 IEEE/CVF
International Conference on Computer Vision Workshop (ICCVW), Oct. 2019, pp. 37–45, doi: 10.1109/ICCVW.2019.00011.
[49] G. Jocher et al., “ultralytics/yolov5: v6.2 - YOLOv5 Classification Models, Apple M1, Reproducibility, ClearML and Deci.ai
integrations,” Aug. 2022, doi: 10.5281/ZENODO.7002879.
[50] D. Du et al., “VisDrone-DET2019: The Vision Meets Drone Object Detection in Image Challenge Results,” in 2019 IEEE/CVF
International Conference on Computer Vision Workshop (ICCVW), Oct. 2019, pp. 213–226, doi: 10.1109/ICCVW.2019.00030.
[51] R. Padilla, S. L. Netto, and E. A. B. da Silva, “A Survey on Performance Metrics for Object-Detection Algorithms,” in 2020
International Conference on Systems, Signals and Image Processing (IWSSIP), Jul. 2020, vol. 2020-July, pp. 237–242, doi:
10.1109/IWSSIP48289.2020.9145130.

407
BIOGRAPHIES OF AUTHORS
Wahyu Pebrianto Wahyu Pebrianto received his bachelor degree in Department
of Information Technology Politeknik Negeri Jember in 2020. Now he is pursuing Master
degree in Universitas Brawijaya. His main research focus is deep learning, computer vision,
and pattern recognition. He can be contacted at email: wahyu.pebrianto1@gmail.com.
Panca Mudjirahardjo Panca Mudjirahardjo received the B.Eng. Degree in
Electrical Engineering from Universitas Brawijaya, Indonesia, in 1995. M.Eng. Degree in
Electrical Engineering from Universitas Gadjah Mada, Indonesia, in 2001 and Dr. Eng. degree
in Control Engineering (Machine Intelligence Lab.) from Kyushu Institute of Technology,
Japan, in 2015. He joined as Assistant Engineer of Production Engineering Dept. at PT. Asahi
Electronics Indonesia, a Telephone Answering Device manufacturer, in 1995-1998; as
Engineering Assistant of Engineering Dept. at PT. Tokyo Pigeon Indonesia, an Audio
Mechanism manufacturer, in 1998-1999. Currently he is with Electrical Engineering Dept. at
Universitas Brawijaya, since 2002. His current research interests include Digital and Analog
Instrumentation System Design, Pattern Recognition, Image Processing and Computer Vision.
He can be contacted at email: panca@ub.ac.id.
Sholeh Hadi Pramono Sholeh Hadi Pramono was borned in 1958. He obtained
his bachelor degree in 1985 majoring in electrical power system from Universitas Brawijaya.
In 1990 and 2009 He obtained his Master and Doctoral degree repectively from Universitas
Indonesia majoring in Optoelectrotechniques and laser application. His major research is in
optical communication, photovoltaic and artificial intelligence. He has been a lecturer in
Universitas Brawijaya since 1986. He can be contacted at email: sholehpramono@ub.ac.id.

Partial half fine-tuning for object detection with unmanned aerial vehicles

More Related Content

Similar to Partial half fine-tuning for object detection with unmanned aerial vehicles (20)

More from IAESIJAI (20)

Recently uploaded (20)

Partial half fine-tuning for object detection with unmanned aerial vehicles