Freelancer logo How It Works Browse Jobs
Log In Sign Up
Post a Project
Profile cover photo
You're now following
Error following user.
This user does not allow users to follow them.
You are already following this user.
Your membership plan only allows 0 follows. Upgrade here.
Successfully unfollowed
Error unfollowing user.
You have successfully recommended
Error recommending user.
Something went wrong. Please refresh the page and try again.
Email successfully verified.
User Avatar
$7 USD / hour
Flag of INDIA
mandi, india
$7 USD / hour
It's currently 4:29 AM here
Joined January 16, 2020
0 Recommendations

Vivek M.

@imvkmittal14

 0.0 (0 reviews)
0.0
0.0
0%
0%
$7 USD / hour
Flag of INDIA
mandi, india
$7 USD / hour
N/A
Jobs Completed
N/A
On Budget
N/A
On Time
N/A
Repeat Hire Rate

Machine Learning & Deep Learning Engineer

I am deep learning engineer. I make models that can beat the human level performance in various tasks. I love to teach machines to do work for us. Experienced in training Neural Networks, Convolutional Neural Networks, Recurrent Neural Networks. I have done various projects involving Computer Vision, Natural Language Processing. Work with me you will get to know me more. My toolbox: 1. Scikit-Learn 2. TensorFlow & Keras 3. PyTorch 4. Pandas 5. Numpy 6. Matplotlib 7. Django 9. CV2 10. SpaCy
Freelancer Machine Learning Experts India
See all Machine Learning Experts in India

Contact Vivek M. about your job

Log in to discuss any details over chat.

Portfolio

9774036
Used: YOLO, Bi-LSTM Architecture & Attention mechanism We propose Attention Grounder (AttnGrounder), a single-stage end-to-end trainable model for the task of visual grounding. Visual grounding aims to localize a specific object in an image based on a given natural language text query. Unlike previous methods that use the same text representation for every image region, we use a visual-text attention module that relates each word in the given query with every region in the corresponding image for constructing a region-dependent text representation. Furthermore, for improving the localization ability of our model, we use our visual-text attention module to generate an attention mask around the referred object. The attention mask is trained as an auxiliary task using a rectangular mask generated with the provided ground-truth coordinates. We evaluate AttnGrounder on the Talk2Car dataset and show an improvement of 3.26% over the existing methods.
Visual Grounding - Detecting objects using Language
Used: YOLO, Bi-LSTM Architecture & Attention mechanism We propose Attention Grounder (AttnGrounder), a single-stage end-to-end trainable model for the task of visual grounding. Visual grounding aims to localize a specific object in an image based on a given natural language text query. Unlike previous methods that use the same text representation for every image region, we use a visual-text attention module that relates each word in the given query with every region in the corresponding image for constructing a region-dependent text representation. Furthermore, for improving the localization ability of our model, we use our visual-text attention module to generate an attention mask around the referred object. The attention mask is trained as an auxiliary task using a rectangular mask generated with the provided ground-truth coordinates. We evaluate AttnGrounder on the Talk2Car dataset and show an improvement of 3.26% over the existing methods.
Visual Grounding - Detecting objects using Language
9773942
Implemented a research paper: Scene Classification With Recurrent Attention of VHR Remote Sensing Images Scene Classification With Recurrent Attention Models of VHR Remote Sensing Images. Used convolutional neural networks and recurrent neural networks. See https://github.com/vk-mittal14/Scene-Classification
Satellite Image Classification
9773920
Archaeologists usually find broken pieces of rare historic items after that they need to arrange the pieces in a particular order that makes some sense but the human mind is somewhat biased about the things that we see usually in our life thus in this project we aim to make a data-driven machine learning model that can solve 3*3 jigsaw puzzles. For this, we use a standard CNN model for extracting features from different regions of the image. After that, the Transformer model is used to relate distinct image regions with each other. Finally, the output of the Transformer is passed to an MLP layer that predicts the correct position for the image regions.
Jigsaw Solver using Neural Networks & Transformers
Archaeologists usually find broken pieces of rare historic items after that they need to arrange the pieces in a particular order that makes some sense but the human mind is somewhat biased about the things that we see usually in our life thus in this project we aim to make a data-driven machine learning model that can solve 3*3 jigsaw puzzles. For this, we use a standard CNN model for extracting features from different regions of the image. After that, the Transformer model is used to relate distinct image regions with each other. Finally, the output of the Transformer is passed to an MLP layer that predicts the correct position for the image regions.
Jigsaw Solver using Neural Networks & Transformers
9773914
I designed a biometric system that takes the face as input and compares it with an already existing image of the owner to predict whether to allow access or not. As the person is wearing the matching complete faces is difficult. Therefore, we match the Forehead Wrinkles to predict whether two images are the same or not. Our pipeline consists of a forehand detection module and a matching module. For forehead detection, we finetune a pretrained Faster-RCNN model, and for forehead matching, we finetune a pretrained ResNet 50 with ArcFaceLoss.
Facial Biometric System using Forehead
9773907
Achieved 2nd position in a Deep Learning competition organized by Facebook AI Research. In this challenge, our goal was to understand the spatial-temporal movement of eye gaze direction. Specifically, we need to predict the gaze direction for future time-steps based on the previously estimated gaze vectors. To this end, we use a two-stage approach in which we use: a ResNet18 backbone for predicting the gaze direction for an image and an LSTM model to capture the temporal dependency between the gaze direction and predict the future gaze direction.
EyeGaze Detection and Future Prediction
9773902
Deepfake Detection
9774036
Used: YOLO, Bi-LSTM Architecture & Attention mechanism We propose Attention Grounder (AttnGrounder), a single-stage end-to-end trainable model for the task of visual grounding. Visual grounding aims to localize a specific object in an image based on a given natural language text query. Unlike previous methods that use the same text representation for every image region, we use a visual-text attention module that relates each word in the given query with every region in the corresponding image for constructing a region-dependent text representation. Furthermore, for improving the localization ability of our model, we use our visual-text attention module to generate an attention mask around the referred object. The attention mask is trained as an auxiliary task using a rectangular mask generated with the provided ground-truth coordinates. We evaluate AttnGrounder on the Talk2Car dataset and show an improvement of 3.26% over the existing methods.
Visual Grounding - Detecting objects using Language
Used: YOLO, Bi-LSTM Architecture & Attention mechanism We propose Attention Grounder (AttnGrounder), a single-stage end-to-end trainable model for the task of visual grounding. Visual grounding aims to localize a specific object in an image based on a given natural language text query. Unlike previous methods that use the same text representation for every image region, we use a visual-text attention module that relates each word in the given query with every region in the corresponding image for constructing a region-dependent text representation. Furthermore, for improving the localization ability of our model, we use our visual-text attention module to generate an attention mask around the referred object. The attention mask is trained as an auxiliary task using a rectangular mask generated with the provided ground-truth coordinates. We evaluate AttnGrounder on the Talk2Car dataset and show an improvement of 3.26% over the existing methods.
Visual Grounding - Detecting objects using Language
9773942
Implemented a research paper: Scene Classification With Recurrent Attention of VHR Remote Sensing Images Scene Classification With Recurrent Attention Models of VHR Remote Sensing Images. Used convolutional neural networks and recurrent neural networks. See https://github.com/vk-mittal14/Scene-Classification
Satellite Image Classification
9773920
Archaeologists usually find broken pieces of rare historic items after that they need to arrange the pieces in a particular order that makes some sense but the human mind is somewhat biased about the things that we see usually in our life thus in this project we aim to make a data-driven machine learning model that can solve 3*3 jigsaw puzzles. For this, we use a standard CNN model for extracting features from different regions of the image. After that, the Transformer model is used to relate distinct image regions with each other. Finally, the output of the Transformer is passed to an MLP layer that predicts the correct position for the image regions.
Jigsaw Solver using Neural Networks & Transformers
Archaeologists usually find broken pieces of rare historic items after that they need to arrange the pieces in a particular order that makes some sense but the human mind is somewhat biased about the things that we see usually in our life thus in this project we aim to make a data-driven machine learning model that can solve 3*3 jigsaw puzzles. For this, we use a standard CNN model for extracting features from different regions of the image. After that, the Transformer model is used to relate distinct image regions with each other. Finally, the output of the Transformer is passed to an MLP layer that predicts the correct position for the image regions.
Jigsaw Solver using Neural Networks & Transformers
9773914
I designed a biometric system that takes the face as input and compares it with an already existing image of the owner to predict whether to allow access or not. As the person is wearing the matching complete faces is difficult. Therefore, we match the Forehead Wrinkles to predict whether two images are the same or not. Our pipeline consists of a forehand detection module and a matching module. For forehead detection, we finetune a pretrained Faster-RCNN model, and for forehead matching, we finetune a pretrained ResNet 50 with ArcFaceLoss.
Facial Biometric System using Forehead
9773907
Achieved 2nd position in a Deep Learning competition organized by Facebook AI Research. In this challenge, our goal was to understand the spatial-temporal movement of eye gaze direction. Specifically, we need to predict the gaze direction for future time-steps based on the previously estimated gaze vectors. To this end, we use a two-stage approach in which we use: a ResNet18 backbone for predicting the gaze direction for an image and an LSTM model to capture the temporal dependency between the gaze direction and predict the future gaze direction.
EyeGaze Detection and Future Prediction
9773902
Deepfake Detection

Reviews

Changes saved
No reviews to see here!

Experience

Research Engineer Intern

Siemens
Jun 2021 - Aug 2021 (2 months, 2 days)
•Developed a prototype algorithm for a mistake-proofing system that uses Visual and Text feedback for assisting workers in industrial settings. •Performed research on multi-modal Vision Transformer (ViT) models. •Designed a novel combination of Transformer models for Scene Understanding using multi-modal inputs. •Extended the code using Horovod for large-scale distributed training with multiple GPUs.

Student Researcher

MIDAS Lab, IIIT Delhi
Jan 2021 - Jun 2021 (5 months, 1 day)
•Developed a novel multi-modal reinforcement learning algorithm for neural quantitative trading based on natural language data. •Developed a novel Multimodal Multivariate Hierarchical Time-Aware Entropy-based Curriculum Learning framework for improving the training of Neural Trading Models. Published at UAI 2021.

Research Assistant

MANAS Lab, IIT Mandi
Dec 2020 - Feb 2021 (2 months, 1 day)
• Worked on Autism Classification using MRI Signals. • Used Vision Transformers Architecture for learning Spatio-temporal relations in MRI signals. • Explored various combinations of neural architecture using CNN, LSTM, and Transformers

Education

Bachelor of Technology in Electrical Engineering

Indian Institute of Technology, India 2018 - 2022
(4 years)

Qualifications

Structuring Machine Learning Projects

Coursera
2020
Learned how to Structure and Develop Machine Learning Projects Certificate: https://coursera.org/share/d96ed88b1809c6d6f27c7bde13a2119f

Sequence Models

Coursera
2020
Learned how to use RNN, GRU, and LSTM. Learned about Word2Vec, GloVE embedding, etc. Certificate: https://coursera.org/share/cf1cba0663dfaa5aabe6505c2bdd902c

Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization

Coursera
2020
Learning about Improving Deep Neural Networks using Hyperparameter Tuning, Regularization and Optimization Certificate: https://coursera.org/share/c83a63c0b9c72e6d4d42f6dd964ac695

Publications

Modeling Financial Uncertainty with Multivariate Temporal Entropy-based Curriculums

Conference on Uncertainty in Artificial Intelligence (UAI), 2021
We propose a novel multimodal curriculum learning approach: FinCLASS, which evaluates stock affecting signals via entropy-based heuristics and measures their linguistic and price-based complexities in a time-aware, hierarchical fashion. We show that training financial models can benefit by exposing neural networks to easier examples of stock affecting signals early during the training phase, before introducing samples having more complex linguistic and price-based temporal variations.

AttnGrounder: Talking to cars with attention

ECCV’20 C4AV Workshop, 2020.
We propose Attention Grounder (AttnGrounder), a single-stage end-to-end trainable model for the task of visual grounding. Visual grounding aims to localize a specific object in an image based on a given natural language text query. Code: https://github.com/vk-mittal14/AttnGrounder

Contact Vivek M. about your job

Log in to discuss any details over chat.

Verifications

Preferred Freelancer
Identity Verified
Payment Verified
Phone Verified
Email Verified
Facebook Connected

Top Skills

Machine Learning (ML)
Natural Language
Web Development
Computer Vision
Deep Learning

Browse Similar Freelancers

Machine Learning Experts in India
Machine Learning Experts
Linguists
Web Developers

Browse Similar Showcases

Machine Learning (ML)
Natural Language
Web Development
Computer Vision
Previous User Next User
Invite sent successfully!
Thanks! We’ve emailed you a link to claim your free credit.
Something went wrong while sending your email. Please try again.
Freelancer logo New Zealand / English Help & Support Freelancer Categories Projects Contests Freelancers Enterprise Membership Preferred Freelancer Program Project Management Local Jobs Photo Anywhere Showcase API for Developers Get Verified Desktop App About About us How it Works Security Investor Sitemap Stories News Team Awards Press Releases Careers Terms Privacy Policy Terms and Conditions Copyright Policy Code of Conduct Fees and Charges Partners Escrow.com Loadshift Warrior Forum Apps Apple App Store logoGoogle Play logo
Copy to clipboard failed, please try again after adjusting your permissions.
Copied to clipboard.
Registered Users Total Jobs Posted
Freelancer ® is a registered Trademark of Freelancer Technology Pty Limited (ACN 142 189 759)
Copyright © 2024 Freelancer Technology Pty Limited (ACN 142 189 759)
Loading preview
Permission granted for Geolocation.
Your login session has expired and you have been logged out. Please log in again.