Skip to main content

Single Shot MultiBox Detector - SSD

Project description

Single Shot Multibox Detector - Facade Parsing

Semester Project at Swiss Data Science Center @EPFL


This project is a collaboration between the Civil Engineers and the Swiss Data Science Center(SDSC) at EPFL. The motivation behind the project is to help Civil Engineers in detecting the damage imposed on buildings by an earthquake. In this semester project, only a sub-part of the whole project has been tackled. By using deep learning, we automate the detection of important objects in an given image. As a deep learning method we are using the Single Shot MultiBox Detector.

Find more details about the project in the report.pdf.


To install the ssd_project library and use it within your python environment:

pip install ssd-project==1.0

Examples on how to use

Examples for various functions and sub-tasks of the project can be found in the Notebooks folder.


To train, the script is ready to use:

usage: [-h] [--epochs EPOCHS] [--split-seed SPLIT_RATIO]
                [--batch-train BATCH_SIZE] [--lr LR]
                [--model PRETRAINED_MODEL] [--path_imgs PATH_IMGS]
                [--path_bboxes PATH_BBOXES] [--path_labels PATH_LABELS]

Train an SSD model

optional arguments:
  -h, --help            show this help message and exit
  --epochs EPOCHS
  --split-seed SPLIT_RATIO
  --batch-train BATCH_SIZE
  --lr LR
  --path_imgs PATH_IMGS
  --path_bboxes PATH_BBOXES
  --path_labels PATH_LABELS

For example:

python3 --epochs=500
python --model saved_models/BEST_model_ssd300.pth.tar

For loading a pretrained a model python --model saved_models/BEST_model_ssd300.pth.tar

When training a new model, the script saves a specific structure, therefore to continue training from the best specific point, just give as input to the saved structure.


For predictions, one should look into notebooks/results.ipynb

Ground Truth & Aspect Ratios

notebooks/creation_ground_truth.ipynb shows how the ground truth was derived. notebooks/Aspect_Ratios.ipynb shows how aspect ratios and scales for prior box are derived


notebooks/Data Augmentations Examples.ipynb shows an example of all the transformations applied to one image.


The best model is saved in saved_model/first_best_model_ssd300.pth.tar

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

ssd_project-1.1-py2-none-any.whl (22.8 kB view hashes)

Uploaded py2

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page