Advanced 3D Neural Architectures — Course Syllabus

Name: Advanced 3D Neural Architectures
Price: 397 EUR

Reference syllabus for the Advanced 3D Neural Architectures course delivered by the 3D Geodata Academy. It defines the learning objectives, audience, technical requirements, the module-by-module program, the assessment scheme, the results indicators and the legal terms of purchase.

"Deploy PointNet++, KPConv and GrowSP on real aerial LiDAR — the system-design thinking that ships production 3D models."

1. Course Overview

Dimension	Details
Format	Self-paced online course delivered through the 3D Geodata Academy LMS.
Price	€397 (excl. VAT). See section 7 for the legal payment terms.
Learning Objectives	Design 3D DL systems: Pick the right architecture, training budget and evaluation protocol before writing a line of code. (M1) Deploy SOTA point networks: Install, train and infer with KPConv and PointNet++ on real outdoor scans. (M3, M4) Ship a production pipeline: Build voxel-based 3D CNN/R-CNN, run the GrowSP pipeline and package a deployable Python app. (M2, M5)
Target Audience	Engineers who have already trained a PointNet and can read PyTorch training loops, ready to deploy PointNet++, KPConv, GrowSP and 3D CNN/R-CNN on real aerial LiDAR.
Prerequisites	Working Python notions help. Watch the prerequisites primer →
Estimated Duration	Approximately 18 hours of focused work. Fully asynchronous.
Access	Direct enrolment via the 3D Geodata Academy. A 14-day legal cooling-off period applies.
Accessibility & Disability	All courses are open to learners with disabilities. A dedicated referent reviews each request to put the right pedagogical and technical adjustments in place. Referent: Dr. Florent Poux — howto@learngeodata.eu.
Contact	Dr. Florent Poux — howto@learngeodata.eu 3D Geodata Academy

A note from Dr. Florent PouxAfter PointNet, most courses stop and most students hit a wall. I built this course for that exact moment: the aerial LiDAR scan with 100M points and five imbalanced classes that PointNet cannot touch. Every architecture in here is one I have deployed on Fortune 500 projects.

2. Technical Stack & Pedagogical Means

Software stack: Python and the standard 3D ecosystem (Open3D, Laspy, NumPy, PyTorch, CloudCompare, Meshroom — adapted to the course focus).
Hardware: any modern laptop (Windows, macOS, Linux). GPU recommended for deep learning modules.
Datasets: reference datasets and notebooks provided.
Infrastructure: proprietary LMS with 24/7 access, automated quizzes, progress tracking and digital course materials.
Modality: alternates theory and hands-on practice — code, real datasets, software use and shippable project deliverables.
Theory resources: video lessons, written handouts, curated research articles and PhD-level references.

3. Course Structure

Module	Title & Focus
M1	3D Deep Learning System Design Architecture selection, data strategy and evaluation plan before the first line of code.
M2	Voxel-based 3D CNN & R-CNN Grid-based architectures built from scratch, compared to point-based alternatives.
M3	KPConv Real-world Application Install, configure, train and infer with the sharpest 3D convolution operator on real scans.
M4	PointNet++ on Aerial LiDAR Hierarchical networks deployed end to end on large outdoor scenes.
M5	GrowSP & Production Deployment The full path from GrowSP pretraining to a deployable Python app.

Why this structure, Dr. Florent PouxEach of the 5 modules ends with a quiz, and the quizzes are cumulative. Don't skip a module just because you think you know it. The gaps you didn't know you had show up in the final quiz.

M1 — 3D Deep Learning System Design

Architecture selection, data strategy and evaluation plan before the first line of code.

3D deep learning system design thinking.
Architecture selection framework.
Training budget planning.
Evaluation protocols.

M2 — Voxel-based 3D CNN & R-CNN

Grid-based architectures built from scratch, compared to point-based alternatives.

Creating a Voxel custom Dataset class (PyTorch).
Creating a 3D CNN for semantic segmentation.
Creating a 3D R-CNN for semantic segmentation.
Tradeoffs vs point-based methods.

M3 — KPConv Real-world Application

Install, configure, train and infer with the sharpest 3D convolution operator on real scans.

KPConv introduction and installation.
Code explanation and configuration.
Training on real data.
Inference and evaluation.

Mid-course checkpoint, Dr. Florent PouxWhen you reach M3 — KPConv Real-world Application, stop and apply what you've learned to a dataset you actually care about. The back half of the course goes faster when the first half sits on a real example, not a toy one.

M4 — PointNet++ on Aerial LiDAR

Hierarchical networks deployed end to end on large outdoor scenes.

PointNet++ functions definition.
Architecture implementation.
Testing and inference.
Aerial LiDAR specifics.

M5 — GrowSP & Production Deployment

The full path from GrowSP pretraining to a deployable Python app.

GrowSP fundamentals and setup.
Code and data structure.
Training, inference and results.
Workflow and system design for production.
Step-by-step 3D Python app production.
Deployable project structure.

Expert tip — Dr. Florent PouxDo not try to run all three networks in parallel. Pick KPConv first, get it green on your own data, then branch into PointNet++ and GrowSP. Sequential debugging beats heroic debugging every time.

4. Assessment, Certificate & Grading

This is a standalone course: there is no project to defend and no oral examination. Evaluation is fully quiz-based, automated through the LMS.

Stage	Activity	Validation
Before the course	Optional positioning quiz to calibrate prior knowledge.	Informative — no minimum score.
During the course	End-of-module quiz (one per module, 10 to 15 questions).	Score ≥ 70 % per quiz.
End of the course	Final quiz covering all modules.	Score ≥ 80 %.

Conditions to obtain the certificate

Validate every end-of-module quiz with a score of 70 % or more.
Validate the final quiz with a score of 80 % or more.
Quizzes can be retaken without limit; the latest score is retained.

Grading scale

Pass: all quizzes ≥ 70 % and final ≥ 80 %.
Pass with Merit: average across all quizzes ≥ 85 %.
Pass with Distinction: average across all quizzes ≥ 92 %.

Successful learners receive the course certificate (PDF + verifiable digital badge) and join the Alumni registry.

Accessibility & disability: all evaluations can be adapted (extended time, alternative formats, oral or written substitution, screen-reader friendly versions) on request to the disability referent howto@learngeodata.eu.

5. Course Results & Quality Indicators

3D Geodata Academy publishes its course performance indicators transparently. Figures below cover this course and are updated at the end of each session.

Indicator	Current Result	Target
Number of enrolled learners	Data being consolidated	Continuous growth
Satisfaction rate	Data being consolidated	> 95 %
Success rate (certificate obtained)	Data being consolidated	> 85 %
Drop-out / interruption rate	Data being consolidated	< 5 %
Recommendation rate	Data being consolidated	> 90 %

Indicators consolidated from in-LMS quizzes and end-of-course satisfaction surveys. Last update: April 2026.

6. Next Step

This course gives you the operational base. To go further with structured mentorship and a wider curriculum, secure your spot below or join the 3D AI Accelerator.

🛒 Secure your spot 🚀 Join the 3D AI Accelerator

The 3D AI Accelerator adds direct mentorship with Dr. Florent Poux, full access to the complete course library (20+ courses), monthly analytics on the 3D spatial AI ecosystem, curated research papers and the private job board with reviews and notes on which roles are worth pursuing.

7. Legal, Accessibility & Purchase Conditions

Digital accessibility

3D Geodata Academy is committed to making its content accessible to learners with disabilities. A dedicated referent oversees content accessibility (captioning, transcripts, screen-reader friendly layouts, alternative evaluation formats). Contact: howto@learngeodata.eu.

Access for learners with disabilities

Every course can be adapted upon request. A short questionnaire at enrolment captures the pedagogical and technical adjustments needed. Disability referent: Dr. Florent Poux — howto@learngeodata.eu.

Payment terms — IP-protection caution & 14-day retraction

Because course content is delivered digitally and immediately accessible, the purchase is structured as follows to protect the intellectual property while preserving your legal right of retraction:

At checkout, the payment is collected as a caution / security deposit for IP infringement protection. Access to the LMS is granted immediately.
The deposit is held for 14 calendar days, matching the legal cooling-off period for distance contracts.
If you exercise your right of retraction within those 14 days and have not consumed a substantial part of the content, the deposit is refunded in full.
At the expiry of the 14-day period, in the absence of a retraction request, the deposit is automatically converted into the actual payment.

By purchasing, the learner acknowledges that the course materials are original works protected by copyright. Any reproduction, redistribution or commercial reuse without prior written consent is prohibited.

Data protection (GDPR)

Personal data is processed for the sole purpose of delivering the course, tracking progress, issuing the certificate and providing customer support. Access, rectification or deletion: howto@learngeodata.eu.

General terms of sale

Full general terms of sale are available on request and joined to every order confirmation.