AI Pallet Box Counter
Budget: $750 – $1,500 USD
I want to automate our warehouse inventory by letting a vision model do the counting for me. The idea is simple: I point a camera at a pallet—whether it is still factory-sealed or half emptied—and your algorithm instantly tells me how many boxes remain. The source can be a single high-resolution photograph or a short video; both formats are common in our operation, so the solution must accept either without extra conversion steps.
The pallets themselves vary. We have the usual 48"×40" units but also a fair share of custom footprints, so the model needs to generalise across both standard and non-standard sizes and cope with irregular stacking patterns.
I am happy with a Python-based approach that leverages modern vision libraries such as PyTorch or TensorFlow and classical helpers like OpenCV. Real-time inference is not mandatory, yet actionable speed on an RTX-class GPU—or even CPU if you can manage it—would be appreciated.
Please deliver:
• A trained model with weights
• A clean inference script that takes an image or video path and returns the box count
• A short README explaining installation, usage, and how to fine-tune on new data
I will validate by running the script on a mixed set of unseen, high-resolution photos and videos of both full and partially emptied pallets; an accuracy above 95 % will be considered a pass.
The pallets themselves vary. We have the usual 48"×40" units but also a fair share of custom footprints, so the model needs to generalise across both standard and non-standard sizes and cope with irregular stacking patterns.
I am happy with a Python-based approach that leverages modern vision libraries such as PyTorch or TensorFlow and classical helpers like OpenCV. Real-time inference is not mandatory, yet actionable speed on an RTX-class GPU—or even CPU if you can manage it—would be appreciated.
Please deliver:
• A trained model with weights
• A clean inference script that takes an image or video path and returns the box count
• A short README explaining installation, usage, and how to fine-tune on new data
I will validate by running the script on a mixed set of unseen, high-resolution photos and videos of both full and partially emptied pallets; an accuracy above 95 % will be considered a pass.