autoresearch-ui-detr-v1

UI element detection fine-tune of facebook/detr-resnet-50 on mrtoy/mobile-ui-design (4 target classes remapped from the source schema).

Track

Track: ui_detr (Gate 2)
Config: 8 epochs, batch 4, grad-accum 4, fp16, T4-class hardware
Classes: rectangle, text, image, group
Train data: 2000-row slice of mrtoy/mobile-ui-design, 10% held out (fixed seed 42)
Training platform: Colab T4 (ZeroGPU free quota is 5 min/day — cannot train; Space is serving-only)

Acceptance gate

eval_map >= 0.35 on the held-out set, no class collapses to zero AP (per-class mAP reported by autoresearch-local/evals/eval_ui_detr.py).
Status: training pending — run train_ui_detr.py (Colab T4 notebook tracks/ui_detr_colab.ipynb), then eval_ui_detr.py.

Usage

from transformers import DetrImageProcessor, DetrForObjectDetection
proc = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50", revision="no_timm")
model = DetrForObjectDetection.from_pretrained(
    "40Hz/autoresearch-ui-detr-v1",
    id2label={0:"rectangle",1:"text",2:"image",3:"group"},
    label2id={"rectangle":0,"text":1,"image":2,"group":3},
    ignore_mismatched_sizes=True)

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for 40Hz/autoresearch-ui-detr-v1

Base model

facebook/detr-resnet-50

Finetuned

(807)

this model