2. ABSTRACT
Accurate object recognition is of great importance in our lives.
Everything from fun apps to industries use object recognition for
many different reasons.
Deep learning can enable us to make robust object detection
algorithms that can “learn” over time making it better every time its
used.
The Aim is to implement the ImageNet dataset into a Simple app that
can recognise objects with good accuracy without the need for
Internet or heavy processing.
3. DEEP LEARNING
Deep Learning is a subfield of machine learning concerned with
algorithms inspired by the structure and function of the brain called
artificial neural networks.
Ordinary Algorithms use large amount of “rules” and require heavy
computation. If the data set increases , the Algorithm only gets
slower.
Deep Learning on the other hand only becomes better performing
with a large dataset and the algorithm usually gets refined to its
simplest form.
4.
5. EXISTING SYSTEMS
Most existing systems used widely these days still need to be
connected to the internet. This is usually done for high accuracy.
But for most applications that level of accuracy is not required. The
app made using ImageNet dataset is up to 80% accurate.
Most existing systems are also not open source making them useless
for the everyday person if S/He does not want to pay a premium.
6. PROPOSED SYSTEM
Using the existing ImageNet data set to build an app that can
continue “Learning”
Optimizing the algorithm to work without internet and on less
powerful hardware.
Detecting multiple objects in the same frame
Giving out multiple probabilities if uncertain about type of object
7. IMAGENET LARGE SCALE VISUAL
RECOGNITION CHALLENGE
The ImageNet Large Scale Visual Recognition Challenge is a
benchmark in object category classification and detection on
hundreds of object categories and millions of images. The challenge
has been run annually from 2010 to present, attracting participation
from more than fifty institutions.
Dataset construction (Large Scale)
Object localization
Object detection dataset
Large scale evaluation
Methods and Results