Week 06 – Research Paper Study & Adaptation
Dates: 2025-07-08 – 2025-07-14 Internship: AI/ML Intern at SynerSense Pvt. Ltd. Mentor: Praveen Kulkarni Sir
Focus
This week was focused on understanding high-impact research papers related to facial recognition and multimodal learning. The insights helped us benchmark approaches and select the right direction for extending our current work — all within NDA constraints.
Goals for the Week
- Analyze top papers including DeepFace, CelebA, and MS-Celeb-1M
- Identify architectural and training strategies from published work
- Extract reusable code components or libraries
- Build internal notes for adaptation (non-disclosable externally)
Tasks Completed
| Task | Status | Notes |
|---|---|---|
| Reviewed DeepFace CVPR 2014 Paper | ✅ Completed | Extracted high-level architecture and pre-processing |
| Summarized CelebA Dataset Usage & Purpose | ✅ Completed | Documented data splits and key attributes |
| Reviewed MS-Celeb-1M Large-Scale Benchmark | ✅ Completed | Noted benchmark criteria, dataset challenges |
| Matched CLIP/VLM output to internal project pipeline | ✅ Completed | Used vision encoder outputs in internal test framework |
| Maintained private internal analysis notes (NDA) | ✅ Completed | For mentor review, not open-sourced |
Key Learnings
- Paper reading deepens understanding of real-world implementations
- Face verification success hinges on alignment + representation
- Vision-language models (like CLIP) can be modularized efficiently
- Datasets vary in complexity, annotation quality, and usage rights
Problems Faced & Solutions
| Problem | Solution |
|---|---|
| No direct code in papers | Used matching PyPI libraries (e.g., DeepFace) |
| Understanding legacy methods | Referred to official benchmark sites and repos |
| Need to protect internal insights | Maintained internal-only notebooks (under NDA) |
📎 References
Goals for Next Week
- Integrate internal code with vision encoder embeddings
- Begin training internal pipeline with updated feature vectors
- Prepare learnings page with clean summaries (non-sensitive)
“Digging into the foundational papers gave me clarity on the ‘why’ behind popular architectures — this shapes how we build our own smarter, compliant solutions.”