Skip to content

[detection] model training iteration with extended multilingual dataset #1604

New issue
New issue
Open
Task
Open
[detection] model training iteration with extended multilingual dataset#1604
Task
Assignees
felixdittrich92SiddhantBahuguna
Labels
benchmarkThis issue reports a benchmarkframework: pytorchRelated to PyTorch backendtopic: text detectionRelated to the task of text detectiontype: enhancementImprovement
Milestone
1.2.0

Description

@felixdittrich92
felixdittrich92
opened on May 22, 2024

🚀 The feature

Retrain detection models from scratch with the updated augmentation pipeline & multilingual extended detection dataset

  • db_mobilenet_v3_large
  • db_resnet50
  • db_resnet34
  • linknet_resnet18
  • linknet_resnet34
  • linknet_resnet50
  • fast_tiny (can be ported to TF @felixdittrich92 )
  • fast_small (can be ported to TF @felixdittrich92)
  • fast_base (can be ported to TF @felixdittrich92)

args:

--rotation
--save_interval
--epochs 20

benchmarks: @felixdittrich92

goal:

  • Improved results on non latin text
  • Model robustness and improvements on edge cases like:

#1535
#1547
#1562
#1570
#1795

Activity

  1. felixdittrich92
    added
    type: enhancementImprovement
    benchmarkThis issue reports a benchmark
    framework: pytorchRelated to PyTorch backend
    framework: tensorflow
    topic: text detectionRelated to the task of text detection
    on May 22, 2024
  2. felixdittrich92
    assigned
    odulcy-mindee
    on May 22, 2024
  3. felixdittrich92
    added this to the 0.10.0 milestone on Jun 6, 2024
  4. felixdittrich92
    modified the milestones: 0.10.0, 0.11.0 on Oct 10, 2024

  5. felixdittrich92 commented on Dec 5, 2024

    @felixdittrich92
    felixdittrich92
    on Dec 5, 2024
    CollaboratorAuthor

    Additional:

    • non latin text like chinese, cyrillic, thai, arabic -> detection improvements
    • dense "small" text instances
    • large text like newspaper headlines
    • additional elements like bullet points
  6. felixdittrich92
    modified the milestones: 0.11.0, 0.12.0 on Jan 28, 2025
  7. felixdittrich92
    assigned
    felixdittrich92
    and
    SiddhantBahuguna
    and unassigned
    odulcy-mindee
    on Mar 5, 2025
  8. felixdittrich92
    changed the title [-][experimential] [detection] model training iteration with updated augmentation pipeline[/-] [+][detection] model training iteration with extended multilingual dataset[/+] on Mar 5, 2025
  9. felixdittrich92
    removed
    framework: tensorflow
    on Jun 13, 2025
  10. felixdittrich92
    modified the milestones: 0.12.0, 2.0.0 on Jun 13, 2025
  11. felixdittrich92
    added the
    Task
    issue type on Jun 13, 2025
  12. felixdittrich92
    modified the milestones: 2.0.0, 1.2.0 on Aug 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

benchmarkThis issue reports a benchmarkframework: pytorchRelated to PyTorch backendtopic: text detectionRelated to the task of text detectiontype: enhancementImprovement

Type

Task

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions