Assignment_ML

Dhineu is offering an internship assignment for evaluating candidates' skills in Machine Learning across four tracks: Computer Vision, NLP/LLM, Vision-Language Models, and GeoAI. Each track has specific objectives, datasets, expected deliverables, and suggested models, requiring candidates to submit their solutions via a GitHub repository or Drive folder. The evaluation criteria include problem understanding, model choice, data handling, implementation quality, results evaluation, reporting, and potential bonus points for additional features.

Uploaded by

av5699987

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

5 views

Assignment_ML

Uploaded by

av5699987

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 3

Assignment Overview:

At Dhineu, we are building cutting-edge solutions at the intersection of Artificial Intelligence, Geospatial
Analytics, and Enterprise Intelligence. As part of our effort to bring on highly capable interns into our
Machine Learning team, we have designed this assignment to evaluate your ability to solve real-world
problems in the domains of Computer Vision, NLP/LLMs, Vision-Language Models, and GeoAI.
This assignment is designed to assess your:
 Problem-solving approach
 Technical and modeling skills
 Creativity in applying ML techniques
 Code quality and reproducibility
 Ability to communicate your results effectively
You are required to choose one (1) out of the four listed assignment tracks and submit your solution in
the form of a GitHub repository or Drive folder containing code, sample outputs, and a brief write-up.
This is your opportunity to show how you approach open-ended AI problems, and we encourage you to
go beyond the minimum requirements — deploy your models, visualize results, or even identify ethical or
geospatial impacts of your work.

Assignment Options (Select anyone)

1. Computer Vision: Object Detection on Aerial Images
 Objective: Detect and classify objects (e.g., vehicles, buildings, garbage, or flooded areas) in
aerial drone imagery.
 Dataset: Drone data
 Expected Deliverables:
o Trained object detection model
o Sample predictions (bounding boxes and labels)
o Evaluation metrics (Precision, Recall, mAP)
o Short report (1–2 pages) on model choice, dataset usage, preprocessing, and results
 Suggested Models: YOLOv8, Faster R-CNN, ViT-based detectors, or other relevant open-source
models

Copyright © 2025 by Dhineu. Permission needs to be granted to use, modify, and distribute this
document.
2. NLP/LLM: Text Summarization & Geo-Tagged QA Pipeline
 Objective:
o Build a system that summarizes articles or answers questions from
structured/unstructured text .
o Make a table by logging geolocation of each news (rows).
 Dataset: CNN-DailyMail News Text Summarization (train.csv), from this link
https://www.kaggle.com/datasets/gowrishankarp/newspaper-text-summarization-cnn-
dailymail
 Expected Deliverables:
o Working summarization or QA pipeline
o Script to extract and geocode geographic references from text
o Final output CSV with summarized content along with geo-coordinates
o Brief report on model architecture and pipeline design
 Suggested Models: T5, BART, LLaMA-2, GPT-4 API, LangChain, Haystack, or other open-source
frameworks
3. Vision-Language Models (VLMs): Image Captioning
 Objective: Generate text captions for street-view images using vision-language models.
 Dataset: GoPro street view footage from this link https://dhineumlassignment.s3.us-east-
1.amazonaws.com/goPro.zip
 Expected Deliverables:
o Trained captioning model
o Sample predictions (images + generated captions)
o Evaluation using BLEU, CIDEr, METEOR scores
o Short report on approach, architecture used, and results
 Suggested Models: BLIP, CLIP, Flamingo, or similar open-source models
4. GeoAI: Satellite Image Change Detection
 Objective: Detect changes in satellite/aerial images over time.
 Dataset: Dataset contain after and before image along with labels (buildings). Please download
it from here https://dhineumlassignment.s3.us-east-1.amazonaws.com/CDD.zip
 Expected Deliverables:
o Preprocessed data pipeline
o Change detection outputs (before/after images with detected changes)

Copyright © 2025 by Dhineu. Permission needs to be granted to use, modify, and distribute this
document.
o Evaluation metrics or visual inspection outputs
o Brief write-up on methodology and model results
 Suggested Models: Siamese networks, UNet variants, ChangeFormer, or other open-source
change detection models
Submission Guidelines
 Submit a GitHub repo or Google Drive link containing:
o Code (Jupyter Notebook or Python scripts)
o Preprocessed sample dataset (or a script to download)
o Final report (max 3 pages)
o README with setup instructions
Bonus Points For:
 Deploying your model via a simple API (FastAPI/Flask)
 Creating a basic visualization dashboard (Streamlit/Gradio)
 Writing clean, well-documented, modular code
 Demonstrating awareness of ethical AI or bias concerns
Evaluation Criteria:
Your submission will be evaluated based on the following:
 Problem Understanding – Clarity in problem definition and domain relevance
 Model Choice & Justification – Appropriateness and reasoning behind selected approach
 Data Handling – Quality of preprocessing, transformation, and use of dataset
 Implementation & Code Quality – Clean, modular, and reproducible code
 Results & Evaluation – Use of appropriate metrics and insights from results
 Reporting – Clarity and completeness of your write-up
 Bonus Points – Deployment, dashboards, visualization, or ethical/geospatial considerations