What Data for This Category Looks Like
Typical participants in this category include annotation teams, computer vision algorithm teams, and quality inspectors. Daily generated documents include annotation specifications, bounding box examples, quality inspection reports, and rework lists. Data sources primarily consist of image samples, annotation platforms, category dictionaries, and spot check results. Update frequency is daily or per project batch.
Unlike text annotation projects, this category processes visual target position and category information. Core data fields include target bounding box coordinates, category labels, spot check markers, and rework counts. Units used are pixels, category numbers, and integer counts.
What Constraints Do These Characteristics Impose on Permission and Data Isolation
Image samples and bounding box examples are visual sensitive data, requiring strict isolation from unauthorized access. Annotation specifications and quality inspection reports contain project-specific rules and spot check results. Leakage of these materials will impact annotation consistency and project compliance.
The daily or batch update data characteristic requires permission configuration to support dynamic adjustments, preventing cross-batch data confusion. Differences in operation scope across roles require a permission system that matches the division of labor among annotators, quality inspectors, and administrators. Visual data storage must be separated from text data to avoid leakage risks from cross-category data co-mingling.
Configuration Settings
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
teamId | Bind a unique teamId to each annotation batch | Matches the batch grouping logic of annotation projects, isolates annotation data across different batches |
tmbId | Assign independent tmbId to annotation groups, quality inspection groups, and administrators | Differentiates operation scopes for different roles, restricts cross-group data access |
| Role permission configuration | Divide three-level permissions based on annotator, quality inspector, and administrator roles; only open corresponding documents and data for each role | Matches the daily division of labor of annotation teams, restricts access to sensitive documents |
| Data storage isolation | Divide independent storage directories by project batch | Corresponds to the batch update characteristic of image samples, avoids data co-mingling |
| Access log retention | Retain for 30 days | Meets traceability requirements for compliance reviews; duration can be adjusted based on actual measurements |
Three Common Misconfigurations
- Failing to bind the corresponding annotation batch when configuring
teamId, resulting in permission denied errors during call testing. The root cause is that the annotation batch is not associated withteamId, so the permission verification logic cannot identify the legitimate access scope. - Not assigning independent
tmbIdto different roles, allowing annotators to view quality inspectors' rework lists and quality inspection reports. The root cause is thattmbIdis not used to isolate operation permissions across roles, leading to cross-group data access. - Not enabling role permission configuration, allowing intern accounts to access all annotation specification documents. The root cause is that default permissions are not divided by role division of labor, exposing all sensitive content and failing to match the role division logic of annotation teams.
How to Verify Successful Configuration
- Access the FastGPT v4.17.0 backend, view the project list bound to
teamId, confirm that only image samples from the corresponding batch are visible. - Switch between different role accounts, verify that only annotation documents and data under the corresponding
tmbIdare accessible. - Check the data storage directory, confirm that image samples from different batches are stored in independent paths with no cross-batch co-mingling.
- Review the access log module, confirm that operation records are retained for the specified duration, enabling traceability of any role's operational behavior.
All configuration values listed above are common starting points, and should be evaluated against the reader's own samples and operational requirements.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.