Deep Learning For Visual Computing | EveryEng

Skip to main content

Self-paced Advanced

Deep Learning For Visual Computing

4(117)

227 views

FREE

603 min

Anytime

English

227 views

Engineering AcademyLearn Without Limits: Free Engineering Courses

Electronics & Instrumentation

Artificial Intelligence

Lifetime access
Anytime Learning
Learn from Industry Expert
Career Option Guideline

Why enroll

This course helps learners understand deep learning in a simple and practical way. Participants gain hands-on experience with Python and PyTorch. It prepares them for real-world applications in computer vision and AI careers.

Your instructor

Engineering Academy

Engineer

Learn Without Limits: Free Engineering Courses

5+ yrs experience·India

Is this course for you?

You should take this if

You work in Electronics & Instrumentation
You're a Artificial Intelligence / Computer Science professional
You have 3+ years of hands-on experience in this field
You want to build skills in Engineering & Design, Research & Developmnet

You should skip if

You're new to this field with no prior experience
You need a different specialisation outside Artificial Intelligence
You need live interaction with an instructor

Course details

Deep learning is a part of machine learning where computers learn to understand data by building knowledge step by step, from simple patterns to complex ideas. For example, when a machine looks at an image, it first learns basic features like lines, edges, curves, and colors. Next, it combines these features to recognize parts of objects such as faces, trees, or buildings. At higher levels, it learns to identify complete objects like people, animals, or mountains, and finally understands the full meaning of the image, such as recognizing a person standing in front of a mountain. Deep learning teaches machines to automatically learn these features and relationships without being explicitly programmed. This approach is widely used in applications like handwritten character recognition, object detection, image captioning, self-driving cars, and generating synthetic images. This course introduces both the theory and hands-on coding practice in deep learning for visual computing, using Python and PyTorch through well-designed practical exercises based on current technologies.

Source: Deep Learning For Visual Computing - IITKGP [Youtube Channel]

Course suitable for

Key topics covered

Introduction to Visual Computing
Feature Extraction for Visual Computing
Neural Networks for Visual Computing
Introduction to Deep Learning with Neural Networks
Multilayer Perceptron and Deep Neural Networks
Autoencoders for Representation Learning
Stacked Autoencoders
Sparse and Denoising Autoencoders
Learning and Optimization

Course content

The course is readily available, allowing learners to start and complete it at their own pace.

25 lectures10 hr 3 min

Deep Learning for Visual Computing (NPTEL Online Course)
4 min
Introduction to Visual Computing
30 min
Feature Extraction for Visual Computing
27 min
Feature Extraction with Python (Hands on)
30 min
Neural Networks for Visual Computing
24 min
Classification with Perceptron Model (Hands on)
29 min
Introduction to Deep Learning with Neural Networks (Part 1)
25 min
Introduction to Deep Learning with Neural Networks (Part 2)
26 min
Multilayer Perceptron and Deep Neural Networks (Part 1)
25 min
Multilayer Perceptron and Deep Neural Networks (Part 2)
25 min
Classification with Multilayer Perceptron (Hands on)
18 min
Autoencoder for Representation Learning and MLP Initialization
29 min
MNIST handwritten digits classification using auto encoders (Hands on)
25 min
Fashion MNIST classification using auto encoders
28 min
ALL-IDB Classification using auto encoders
24 min
Retinal Vessel Detection using auto encoders (Hands on)
27 min
Stacked Autoencoders
17 min
MNIST and Fashion MNIST Classification with Stacked Autoencoders (Hands on)
27 min
Sparse and Denoising Autoencoders
25 min
Sparse Autoencoders for MNIST classification (Hands on)
26 min
Denoising Autoencoders for MNIST classification (Hands on)
20 min
Cost Function
25 min
Classification cost functions
21 min
Gradient Descent Learning Rule
28 min
SGD and ADAM Learning Rules
18 min

Opportunities that await you!

Skills & tools you'll gain

Engineering & Design

Research & Developmnet

Project Management

Career opportunities

Why people choose EveryEng

Industry-aligned courses, expert training, hands-on learning, recognized certifications, and job opportunities-all in a flexible and supportive environment.

FREE

Access anytime

Questions and Answers

Q: You're validating a perception stack and googling "camera sensor corrosion equivalent in deep learning dataset shift from road salt glare" during winter testing. The training set was collected in dry summer conditions. Which degradation mechanism dominates the model's performance loss?

A: Pick the wrong mechanism and you'll chase retraining knobs while winter ADAS false negatives keep failing track tests. Salt glare and seasonal lighting change the input distribution while the task stays the same, so addressing covariate shift with domain adaptation or targeted data fixes the root cause.

Q: You're tuning a CNN and searching "calculate output feature map size convolution stride padding visual computing" under a build freeze. Input is 224x224, kernel 7x7, stride 2, padding 3. What spatial size exits the conv layer?

A: Get this wrong and the network won't even load, burning a night of CI time. Using (N−K+2P)/S+1 gives (224−7+6)/2+1 = 112, keeping tensor interfaces consistent through the graph.

Q: During a design review you type "estimate cnn inference latency from flops and gpu tflops" before a demo in 30 minutes. A model needs 30 GFLOPs and runs on a 10 TFLOP GPU at 30% utilization. Order-of-magnitude latency per frame?

A: Guess too low and the real-time demo drops frames on stage. 30 GFLOPs at an effective 3 TFLOPs gives about 0.01 seconds, a quick sanity check that matches observed GPU inference behavior.

Q: You're selecting an architecture and searching "vision transformer vs cnn for embedded automotive perception" with ECU thermal limits fixed. Which choice best fits low-latency edge deployment?

A: Choose poorly and the ECU overheats during summer validation runs. Depthwise CNNs keep compute and memory predictable, matching embedded constraints without attention overhead that kills latency.