Mobile Engineering Manager at Groww.

medium

github

productHunt

substack

devto

hashnode

dribbble

gumroad

youtube

Param Aggarwal  • Peerlist

- Highly experimental project out of personal interest
- Successfully visualised the internal layer outputs of my neural network in realtime
- Was able to better understand the network behaviour after visualizing

Attempting to Visualize a Convolutional Neural Network in Realtime

- Collect driving data from simulator
- Run and test different end-to-end learning neural network architectures
- Managed to run the vehicle around the lap with very tiny 32x32 input images and an equally small network, hence being computat

Behavioural Cloning with an Autonomous Car in Simulator

- Process input image using sober operator to detect edges in different ways
- Combine separate detections into one binary image
- Run curve fitting to find polynomial of lane line
- Map the lane image back to input image using perspective 

Detecting Lane Lines using Sobel Operator and Curve Fitting

- Detect lane lines using OpenCV
- Testing on driving data collected from indian roads in Bangalore
- Praise from course instructors for my work
- Project video received upwards of 5000 views

Lane Detection using OpenCV and Testing on Indian Roads

- We had to build an iOS app from scratch
- I helped implement multiple screens and workflows as per design
- Soon after launch app was contributing upwards of 20% revenue

Myntra iOS App (v1 launch)

- Classified thousands of traffic sign images using Convolutional Neural Networks
- Learnt a lot about different architectures by trying upwards of 50 variations and documenting results
- Article was praised by instructors and recommended b

Traffic Sign Classification using Neural Networks

- Used the sliding window approach to pull out pieces of input image
- Extract HOG features from each window
- Distinguish cars and non-cars using linear classifier on each extracted feature window
- Use heat maps to reduce false detections