TimelapseFromPrompt

TimelapseFromPrompt is a tool for automatically generating object-centric timelapse image using natural language prompts.

🔥 About

🔍 What it does

• Input: A regular video recording of any scene (e.g., quadruped, animals, drone).

• Step 1 – Frame Extraction: Uses ffmpeg to extract frames from the video.

• Step 2 – Object Detection: Applies GroundingDINO to detect the object of interest based on a user-defined text prompt.

• Step 3 – Timelapse Creation: Selects and compiles relevant frames into a focused timelapse image of the specified object.

⚙️ Features

• 💬 Prompt-based detection – Just describe what you want to track (e.g., "drone", "person walking", "red ball").
• 🎞️ Frame interval control – Choose how often frames are sampled using frame_interval (default: 13).
• 📦 Detection thresholds – Tune BOXTHRESHOLD and TEXTTHRESHOLD for controlling detection sensitivity:

python BOX_THRESHOLD = 0.35 TEXT_THRESHOLD = 0.25 - 🔁 Dynamic frame interval – Change frame interval mid-video to better highlight interesting segments.

🛠️ Installation

We provide two ways to install the requirements

Normal

1. Use Python3.10 and create a venv ``` sudo apt-get install python3.10-dev python3.10-venv

git clone git@github.com:MaevaGuerrier/TimelapseFromPrompt.git

cd TimelapseFromPrompt/ python3.10 -m venv .venv source .venv/bin/activate pip install torch==2.5.1+cu121 --extra-index-url https://download.pytorch.org/whl/cu121
pip install -e . ```

1. Get weights from the groundingDino repository to run the inference

cd TimelapseFromPrompt/ mkdir weights/ wget -q https://github.com/idea-research/groundingdino/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth

Docker

1. Build and use the dockerfile from the Docker folder

Generating the Timelapse

⭐ Troubleshooting

TODO

✒️ Citation

TimelapseFromPrompt

BibText @software{Guerrier_TimelapseFromPrompt_A_Prompt-Based_2025, author = {Guerrier, Maeva and Soma, Karthik}, license = {MIT}, month = jul, title = {{TimelapseFromPrompt: A Prompt-Based Tool for Object-Focused Timelapse Generation}}, url = {https://github.com/MaevaGuerrier/TimelapseFromPrompt}, version = {1.0.0}, year = {2025} }

Grounding dino

BibTeX @article{liu2023grounding, title={Grounding dino: Marrying dino with grounded pre-training for open-set object detection}, author={Liu, Shilong and Zeng, Zhaoyang and Ren, Tianhe and Li, Feng and Zhang, Hao and Yang, Jie and Li, Chunyuan and Yang, Jianwei and Su, Hang and Zhu, Jun and others}, journal={arXiv preprint arXiv:2303.05499}, year={2023} }