PROJECT DETAILS
AgriDrone Vision Evaluation Pipeline
AgriDrone Vision Evaluation Pipeline is a generalized and anonymized technical case study of an applied computer vision and geospatial machine learning system for agricultural drone imagery.
The project documents the design of an end-to-end ML engineering workflow capable of processing high-resolution UAV images and videos, running object detection models, validating model performance, generating reproducible evaluation artifacts, enriching detections with geospatial metadata, and exporting results into formats suitable for GIS analysis, technical reporting, and downstream agricultural decision-support workflows.
At its core, the system focuses on applying YOLOv8 / YOLO11 object detection models to agricultural imagery captured by drones. The pipeline supports both direct YOLO inference and SAHI-based sliced inference, which is especially useful for high-resolution images where small objects may be lost during resizing. SAHI allows the image to be divided into overlapping slices, processed independently, and reconstructed into full-image detections, improving the ability to detect small or dense objects in aerial scenes.
The documented workflow includes several stages of the machine learning lifecycle: dataset configuration, model training orchestration, validation, benchmarking, automatic best-model selection, inference, post-processing, evaluation, visualization, geospatial export, and technical documentation. It is not presented as a simple model-training notebook, but as a broader ML engineering system that connects experimentation, inference, evaluation, and geospatial analysis.
The training and validation layer describes how YOLO models can be trained and evaluated across different configurations such as model version, image size, batch size, confidence threshold, device selection, and dataset structure. It also documents model-selection logic based on experiment metrics, allowing the system to identify the best available `best.pt` checkpoint from previous runs. The validation and benchmarking workflow includes global metrics, per-class metrics, inference timing, GPU-aware execution considerations, and reproducible artifact generation.
The evaluation layer includes COCO-style evaluation concepts, conversion of YOLO annotations and predictions into COCO-compatible structures, and the calculation of metrics such as AP50, AP50:95, precision, recall, and F1-score. The documentation distinguishes between YOLO-native validation metrics, COCO-style evaluation metrics, and operational inference outputs, making clear that each type of metric answers a different question about model performance.
The inference layer supports processing individual images, full directories of images, high-resolution drone imagery through SAHI, and video inputs. For static imagery, the system generates styled images with bounding boxes, class labels, confidence scores, per-image JSON metadata, batch summaries, and geospatial outputs. For video, the system documents a dedicated video inference and object tracking processor that uses YOLO tracking IDs to count unique objects across frames, render annotated videos, generate JSON summaries, and optionally produce frame-level SRT artifacts.
A major part of the project is the integration between computer vision and geospatial processing. The system does not stop at object detection. It enriches detections with EXIF/GPS metadata when available, converts latitude and longitude into UTM coordinates, and prepares structured spatial outputs for GIS workflows. These outputs may include GeoJSON, CSV, Shapefiles, QGIS-compatible summaries, and spatial metadata records. This allows model predictions to be inspected, analyzed, and visualized in geospatial tools instead of remaining isolated as pixel-space bounding boxes.
The project also documents a raster georeferencing workflow for styled detection images. This includes copying EXIF/XMP metadata from original drone images, generating JGW world files for styled JPEG outputs, documenting CRS assumptions, supporting optional GeoTIFF fallback, and enabling QGIS-oriented raster loading workflows. This raster-oriented layer is separate from vector outputs like GeoJSON and Shapefiles, and it addresses the practical need to visually overlay annotated detection imagery inside GIS tools.
The video processing component introduces a different type of complexity because video inference is not simply image inference repeated over frames. It includes temporal state, frame decoding, YOLO `model.track()` execution, extraction of bounding boxes and tracking IDs, unique object counting, rendering of overlays, video writing, SRT generation, frame-level error handling, and resource finalization. The documentation captures risks such as tracker ID instability, missing `box.id` values, RGB/BGR color-space issues, JSON serialization problems, and performance bottlenecks caused by frame-by-frame processing.
The system also documents important engineering constraints and trade-offs. These include SAHI runtime overhead, GPU/CUDA memory pressure, cuDNN runtime assumptions, filesystem-based experiment lineage, path fragility, lack of formal retry logic, absence of background workers, output idempotency concerns, metadata quality issues, CRS ambiguity, and the difference between a research-grade batch pipeline and a fully production-ready platform. These limitations are intentionally documented to show realistic engineering judgment rather than presenting the system as a finished enterprise product.
From a software architecture perspective, the project is organized around several conceptual services and processing layers: CLI orchestration, model selection, training, validation, benchmarking, inference, SAHI reconstruction, video tracking, geospatial enrichment, raster georeferencing, GIS export, reporting, and documentation rendering. The current public version describes these as generalized architectural components rather than publishing proprietary source code or institutional implementation details.
The repository is intended to demonstrate applied ML engineering capabilities across multiple domains: computer vision, object detection, high-resolution image processing, geospatial metadata handling, GIS integration, video analytics, evaluation methodology, GPU-aware workflows, and technical documentation. It shows how a machine learning pipeline can be extended beyond model inference into reproducible evaluation, spatial intelligence, reporting, and practical analysis workflows.
This public repository is a sanitized portfolio version. It does not include private datasets, source code from proprietary systems, trained model weights, real drone imagery, real field coordinates, real shapefiles, real GeoJSON outputs, production credentials, institutional or client names, unpublished experimental results, or confidential operational details. The documentation is presented as a generalized technical case study focused on architecture, methodology, engineering decisions, and system design patterns.
Framework: AI / Computer Vision
Technology Stack:
Python, PyTorch, Ultralytics YOLOv8/YOLO11, SAHI, OpenCV, Pillow, NumPy, Pandas, pycocotools, ClearML, CUDA, cuDNN, EXIF/GPS metadata processing, UTM coordinate conversion, GeoJSON, Shapefile, JGW world files, GeoTIFF, QGIS, PyQGIS, GDAL/OGR, Markdown documentation
About Team
Company / Institution: Anonymized Agricultural Computer Vision R&D Case Study
Developer: Marco Parra
The repository documents the problem, architecture and results of this case study. Source code is not published (confidential, anonymized professional work).
Project Images
Endpoints & Routes
Available EndPoints
Show Code
views.py Code
@throttle_classes([TenCallsPerMinute])
def projects_rendering(request, project_id):
# Una ficha inexistente devolvía 500 (el .get() lanzaba DoesNotExist): 107 ids
# sin proyecto servían la página desnuda de Django en vez del 404 del sitio.
project = get_object_or_404(Project, id=project_id)
projects = Project.objects.exclude(title_en='My Portfolio').exclude(status='archived')
# image_logo admite NULL (blank=True, null=True) y os.path.basename(None) lanza
# TypeError: una ficha creada en el admin sin logo devolvía 500. Sin logo, la
# plantilla omite la <img>.
logo = extract_filename_from_path(project.image_logo) if project.image_logo else ''
# Galería y endpoints salen de datos, no de código: el prefijo del proyecto
# (slug del repositorio, o el mapa de los proyectos antiguos) y lo que existe en
# static/img y endpoints.json. Publicar un proyecto nuevo es crear la ficha en el
# admin y dejar <slug>1.svg, <slug>2.svg… en static/img.
prefix = gallery_prefix(project)
with open(os.path.join(settings.STATICFILES_DIRS[0], 'json', 'endpoints.json')) as file:
all_endpoints = json.load(file)
# Sin prefijo, startswith('') aceptaría las rutas de todos los proyectos.
endpoints = {key: value for key, value in all_endpoints.items() if prefix and key.startswith(prefix)}
project.website = patch_url(project.website)
project.repository = patch_url(project.repository)
# Artículos del blog sobre este proyecto: uno por artículo, en el idioma activo si
# existe (la misma regla que el listado del blog). Import local: blog ya depende de
# portfolio (Post.project) y así no se forma un ciclo al cargar los módulos.
from blog.models import pick_language
related_posts = pick_language(
project.posts.filter(published_at__lte=timezone.now()), get_language()
)
context = {
'project': project,
'projects': projects,
'related_posts': related_posts,
'path_images': gallery_images(project),
'logo': logo,
'endpoints': endpoints,
'og_image_static': share_image(project),
**show_code([projects_rendering], 'project.html'),
}
return render(request, 'project.html', context)
Template
{% extends 'base.html' %}
{% load i18n static %}
{# Título, descripción e imagen de la ficha: sin esto las diez fichas se compartían con el título genérico del sitio. #}
{% block title %}{{ project.title }} | Marco A. Parra F.{% endblock %}
{% block meta_description %}{{ project.description|striptags|truncatechars:160 }}{% endblock %}
{% block og_title %}{{ project.title }}{% endblock %}
{% block og_description %}{{ project.description|striptags|truncatechars:200 }}{% endblock %}
{% block og_image %}{% if og_image_static %}{{ request.scheme }}://{{ request.get_host }}{% static og_image_static %}{% else %}{{ block.super }}{% endif %}{% endblock %}
{% block twitter_card %}{% if og_image_static %}summary_large_image{% else %}summary{% endif %}{% endblock %}
{% block structured_data %}{{ block.super }}
<script type="application/ld+json">
{"@context": "https://schema.org", "@type": "SoftwareSourceCode", "name": "{{ project.title|escapejs }}", "description": "{{ project.description|striptags|truncatechars:300|escapejs }}", "url": "{{ canonical_url|escapejs }}",{% if project.repository %} "codeRepository": "{{ project.repository|escapejs }}",{% endif %} "programmingLanguage": "{{ project.technology|escapejs }}", "inLanguage": "{{ LANGUAGE_CODE }}", "author": {"@id": "{{ request.scheme }}://{{ request.get_host }}/#person"}}
</script>
{% endblock %}
{% block content %}
{% load static %}
{% load fontawesome_5 %}
{% load cached_pygments %}
<br><br>
<section class="container-section mb-4">
<div class="container-fluid">
<div class="row g-0">
<div class="portfolio-box-caption">
<div class="row">
<div class="col">
<h2 class="portfolio-box-headers" style="text-align:center">
{% trans "PROJECT DETAILS" %}</h2>
</div>
</div>
</div>
<hr>
<div class="container">
<div class="row" >
<div class="container-fluid specialization-detail" style="background: white">
<div class="row mb-4">
<h3 class="mt-9" style="margin:10px">{{ project.title }}</h3>
</div>
<hr>
<div class="row mb-4">
<p class="d-none d-sm-block show-read-more" style="margin:10px">{{ project.description|linebreaksbr }}</p>
</div>
<div class="row mb-4">
<div class="col-12 col-sm-10">
<h4>{% trans "Framework:" %} <strong> {{project.technology|linebreaksbr}}</strong></h4>
</div>
{% if logo %}
<div class="col">
<img class="mw-100" style="height: 3em" src="{% static 'img/' %}{{ logo }}" alt="{{project.title}}" />
</div>
{% endif %}
</div>
<div class="row mb-4">
<hr>
<h6>{% trans "Technology Stack:" %}</h6>
<p style="white-space: pre-wrap;">{{project.technology_stack|linebreaksbr}}</p>
<hr>
</div>
<div class="row mb-4">
<h4>{% trans "About Team" %}</h4>
<div class="row">
<div class="col-12 col-md-6">
<p><strong>Company / Institution: </strong>{{project.organizations}}</p>
</div>
<div class="col-12 col-md-6">
<p><strong>{% trans "Developer:" %} </strong>{{project.developers}}</p>
</div>
</div>
</div>
{% if project.website and project.website != 'Not Available' and project.website != '' %}
<div class="row mb-3">
{% if project.id == 4 or project.id == 9 %}
<div class="col-12 col-sm-6 mb-3 mb-sm-0">
<button class="btn btn-warning rounded-0"><a href="{{ project.website }}" target="_blank" style="font-size: 2em">Server Side Repo</a></button>
</div>
{% else %}
<div class="col-12 col-sm-6 mb-3 mb-sm-0">
<button class="btn btn-warning rounded-0"><a href="{{ project.website }}" target="_blank" style="font-size: 2em">{% trans "Go To Website" %}</a></button>
</div>
{% endif %}
{% if project.id != 2 %}
<div class="col-12 col-sm-6 mb-3 mb-sm-0">
<button class="btn btn-warning rounded-0"><a href="{{ project.repository }}" target="_blank" style="font-size: 2em">Go To Repository</a></button>
</div>
{% endif %}
</div>
{% endif %}
{% if project.id == 1 %}
<div class="row mb-3">
<div class="col-12">
<a class="btn btn-warning rounded-0" href="https://github.com/dekaacevedo/demizona/commits/master/?author=maaferna" target="_blank">
<i class="fab fa-github me-1"></i>My commits in this team project (57)
</a>
<p class="text-muted small mt-2 mb-0">
Team capstone with 6 contributors — the commit history filtered by my authorship
is publicly verifiable in the shared repository.
</p>
</div>
</div>
{% endif %}
{% if project.id == 9 or project.id == 42 or project.id == 75 %}
<div class="row mb-3">
<div class="col-12">
<a class="btn btn-warning rounded-0" href="{{ project.repository }}" target="_blank">
<i class="fab fa-github me-1"></i>Case study repository
</a>
<p class="text-muted small mt-2 mb-0">
The repository documents the problem, architecture and results of this case study.
Source code is not published{% if project.id != 9 %} (confidential, anonymized professional work){% endif %}.
</p>
</div>
</div>
{% elif project.id == 108 or project.id == 111 %}
{% comment %}
Case studies solo documentales sin demo (108 y control dimensional): el
bloque de arriba oculta los botones cuando no hay website.
La condición era «not project.website and project.repository», y capturaba
cualquier ficha sin sitio: la 1 (Demizona) publicaba «No source code is
published» sobre un repositorio de bootcamp que sí tiene código, y con el
botón apuntando al perfil de GitHub, no a un repo. Se enumeran los ids,
como hace la rama de arriba con 9/42/75.
{% endcomment %}
<div class="row mb-3">
<div class="col-12">
<a class="btn btn-warning rounded-0" href="{{ project.repository }}" target="_blank">
<i class="fab fa-github me-1"></i>{% trans "Case study repository" %}
</a>
<p class="text-muted small mt-2 mb-0">
{% trans "Documentation-only repository: problem, method, architecture and limits. No source code or client data is published." %}
</p>
</div>
</div>
{% elif project.repository and not project.website and project.id != 1 %}
{% comment %}
Cualquier ficha nueva sin demo: sin esta rama, la cadena por ids de arriba la
dejaba sin ningún enlace al repositorio. Texto neutro, sin afirmar que no hay
código; la 1 (Demizona) tiene su propio botón a los commits.
{% endcomment %}
<div class="row mb-3">
<div class="col-12">
<a class="btn btn-warning rounded-0" href="{{ project.repository }}" target="_blank">
<i class="fab fa-github me-1"></i>{% trans "Public repository" %}
</a>
</div>
</div>
{% endif %}
{% if project.id == 8 %}
<div class="row mb-3">
<div class="col-12">
<a class="btn btn-warning rounded-0" href="{% url 'todo_demo' %}">
<i class="fas fa-play me-1"></i>{% trans "Try the live demo (no account needed)" %}
</a>
<p class="text-muted small mt-2 mb-0">
{% trans "The running app is the <code>todo_app</code> module of this portfolio (Django + DRF, session auth, AJAX). The linked repository is the original 2023 standalone version (Django REST + Vue), kept archived for reference." %}
</p>
</div>
</div>
{% endif %}
{% if project.id == 2 %}
<div class="row mb-3">
<div class="col-12">
<a class="btn btn-warning rounded-0" href="{% url 'parts' %}">
<i class="fas fa-route me-1"></i>{% trans "Technical deep-dive: routes & endpoints" %}
</a>
</div>
</div>
{% endif %}
</div>
</div>
</div>
</div>
</div>
</section>
{% block certificates %}
{% include 'partials/carousel_images.html' %}
{% endblock %}
{# Artículos del blog sobre este proyecto (Post.project), en el idioma activo si existe. #}
{% if related_posts %}
{% get_current_language as CURRENT_LANGUAGE %}
<section class="container-section mb-4" id="related-articles">
<div class="container-fluid">
<div class="row g-0">
<div class="portfolio-box-caption">
<h2 class="portfolio-box-headers text-center">{% trans "Related articles" %}</h2>
</div>
<hr>
<div class="container mb-4">
<ul class="list-unstyled mb-0">
{% for post in related_posts %}
<li class="py-2 border-bottom">
<a href="{% url 'blog-post-detail' post.slug %}" class="fw-semibold">{{ post.title }}</a>
{% if post.language != CURRENT_LANGUAGE %}<span class="badge bg-secondary align-middle ms-1">{{ post.language|upper }}</span>{% endif %}
<p class="text-muted small mb-0">{{ post.summary|truncatewords:30 }}</p>
</li>
{% endfor %}
</ul>
</div>
</div>
</div>
</section>
{% endif %}
{# Sólo si endpoints.json tiene entradas para este proyecto: un caso de estudio o una CLI no tiene rutas HTTP, y la tabla vacía (fichas 9, 42 y 111) parecía un error. Antes se ocultaba por id, y cada ficha nueva la mostraba vacía. #}
{% if endpoints %}
<section class="container-section mb-4">
<div class="container-fluid">
<div class="row g-0">
<div class="portfolio-box-caption">
<div class="row">
<div class="col">
<h2 class="portfolio-box-headers" style="text-align:center">
<i class="fas fa-route"></i> {% trans "Endpoints & Routes" %}</h2>
</div>
</div>
</div>
<hr>
<div class="container mb-4" id="Endpoint">
<h3 class="mb-3 mt-4 p-4" style="background-color: #212529; color: #fff;"><i class="fas fa-list"> Available EndPoints</i></h3>
{% if project.id == 3 %}
<p>
In this project, only a few available endpoint is displayed, in the section <strong>Available EndPoint</strong> there are displayed. To see the complete code can visit the original repository.
</p>
{% endif %}
<div class="row">
<div class="col table-responsive">
<table class="table table-hover align-middle">
<thead class="table-dark">
<tr>
<th scope="col">{% trans "Endpoint" %}</th>
<th scope="col">{% trans "Description" %}</th>
<th scope="col">URL</th>
</tr>
</thead>
<tbody>
{% for key, values in endpoints.items %}
{% for items in endpoints.values %}
{% for value in items %}
<tr>
<td> {{ value.endpoint }}</td>
<td>{{ value.description|linebreaksbr }}</td>
<td>
<a class="btn btn-success btn-sm rounded-0" href="{{ value.href }}" target="_blank" rel="noopener" aria-label="{% trans "Open" %} {{ value.endpoint }}"><i class="fa fa-edit" aria-hidden="true"></i></a>
</td>
</tr>
{% endfor %}
{% endfor %}
{% endfor %}
</tbody>
</table>
</div>
</div>
</div>
</div>
</div>
</section>
{% endif %}
{% include 'partials/_show_code.html' %}
{% endblock %}