Scribd
Upload
16 views5 pages

Object Ditection Assignment

Uploaded by

Rony
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
16 views5 pages

Object Ditection Assignment

Uploaded by

Rony
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

A comparative study on One-stage vs.

two-
stage object detectors
One-stage object detectors refer to neural networks anticipating all the bounding boxes in a single trip
through the web. Mobile devices are better suited for these tasks due to increased speed and
compatibility. Some of the most often encountered instances of one-stage object detectors include
YOLO, SSD, SqueezeDet, and DetectNet.In contrast, two-stage object detectors employ a two-step
approach. Initially, they utilize region suggestions to create preliminary object proposals. Subsequently,
a specialized per-region head categorizes and enhances these ideas. One-stage detectors typically
exhibit faster processing speeds, while two-stage detectors generally provide superior accuracy.

The prevalent architectures for one-stage object detection include YOLO, SSD, and EfficientDet. The
YOLO (You Only Look Once) system is an object recognition system that operates in real time. It utilizes a
singular neural network to make predictions regarding bounding boxes and class probabilities straight
from whole images in a single evaluation. Single Shot Detection (SSD) is a one-stage object detection
method that uses a solitary deep neural network to immediately estimate bounding boxes and class
probabilities from complete images in a single evaluation. EfficientDet refers to a collection of one-stage
object detectors that employ EfficientNet as the underlying network architecture, demonstrating
exceptional performance on diverse object detection benchmarks.

One-stage object detectors offer several advantages, including enhanced speed and excellent
compatibility with mobile devices. In addition, they necessitate a reduced amount of memory and
computational resources. One of the drawbacks of single-stage object detectors is their relatively lower
accuracy than two-stage detectors. Additionally, these detectors may have challenges in accurately
identifying small items or objects close to each other.

One advantage of two-stage object detectors is their tendency to achieve higher accuracy levels than
one-stage detectors. Additionally, they exhibit enhanced proficiency in identifying things with irregular
shapes or clusters of smaller objects. One of the drawbacks of two-stage detectors is their slower
processing speed and increased need for memory and computational resources compared to one-stage
sensors.

Yolo:

• In YOLO, Object detection as a single regression problem also known as single stage detector

bbox coordinates & class probabilities all are computed in just a single run of algorithm

• YOLO sees the entire image during training & test time so it makes < 1/2 the no. of background
errors compared to Fast R-CNN.
• Intersection Over Union (IOU) is a metric used to judge the accuracy of the bounding box
predicted by the model.
• Non-max Suppression reduces the no. of predicted b.boxes by taking the larges probability
associated with each detection
• Drawbacks:
• The major problem with YOLOv1 is its inability to detect very small objects.

Fast R-CNN:

• Similar to R-CNN, just that it expects region proposals to be fed in with images rather than
proposing them itself
• We run the CNN only once per image sharing the computation among 2,000 region proposals.
• The CNN processes the image and outputs a feature map
• Input region proposals are used to extract the ROI from feature map & create a region proposal
feature map for each proposed region called a the ROI Projection
• Then down-sample this feature map with the help of a ROI pooling layer to get a fixed length
feature map
• Drawbacks:
• much faster as compared to R-CNN. But with large real-life datasets, it does not works so fast
Result:

Here is Fast R-CNN result

Here is Yolov5 result


In general, Faster R-CNN is a good choice for applications where accuracy is critical. YOLOv5 is a good
choice for applications where speed is critical.

Feature Faster R-CNN YOLOv5


Accuracy High Medium
Speed Slow Fast
Memory usage High Medium
Complexity High Medium

Ultimately, the best model for a particular application will depend on the specific requirements of that
application.

In some cases, Faster R-CNN may be the better choice, even if speed is a concern. For example, if the
application is only processing a small number of images, the slower speed of Faster R-CNN may not be a
significant drawback.

In other cases, YOLOv5 may be the better choice, even if accuracy is a concern. For example, if the
application is processing a large number of images in real time, the faster speed of YOLOv5 may be
essential.

The best way to choose between Faster R-CNN and YOLOv5 is to experiment with both models and see
which one works best for the particular application.

Reference:

1. https://www.analyticsvidhya.com/blog/2022/06/yolo-algorithm-for-custom-object-detection/
2. https://viso.ai/deep-learning/object-detection/#:~:text=on%20Viso%20Suite-
,Most%20Popular%20Object%20Detection%20Algorithms,the%20single%2Dshot%20detector%2
0family
3. https://pjreddie.com/darknet/yolo/
4. https://www.v7labs.com/blog/yolo-object-detection

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy