Evaluate one surface-defect detection approach and give operators a simple way to review its mistakes.
Singapore Remote
This is an example brief. Its company, dates, budget, assignment and progress illustrate a project; they do not represent available or commissioned work. Proposals cannot be submitted.
The brief
We have 1,200 labelled photographs of moulded covers from a single camera station. This feasibility package compares a simple image-feature baseline with one compact pretrained model, using a held-out production batch. It includes a local review screen for up to three defect categories and two review sessions. It does not include camera installation, production integration or a guaranteed detection rate. A recommendation against deployment is an acceptable result if the evaluation supports it.
A running start
Completed: reviewed the supplied labels and separated the held-out batch from training inputs.
Current: evaluate the two approaches and connect the selected results to the operator review screen.
Confirm a clean setup reproduces the report, review ten representative errors and hand over the source.
Working together
Confirm the deliverables, review checks and response window together so everyone knows what a successful handover looks like.
Record the requested change, its reason and its effect on deliverables, budget and timing. Both parties agree the revised scope in writing before the additional work begins.