Detectra v3: art/anime/meme/screenshot blind spots closed

README.md

@@ -10,7 +10,7 @@ tags:
 base_model: OwensLab/commfor-model-384
 ---
 
-# Detectra v1
+# Detectra v3
 
 On-device AI-image detector powering the [Detectra Chrome extension](https://github.com/ashhart/Detectra):
 a ViT-S/16 @384 binary classifier (`sigmoid(logit)` = probability the image is
@@ -20,18 +20,23 @@ AI-generated), exported to single-file fp16 ONNX for ONNX Runtime Web
 Fine-tuned from the MIT-licensed
 [Community Forensics ViT-S](https://huggingface.co/OwensLab/commfor-model-384)
 (Park & Owens, CVPR 2025) on modern generators (DALL·E 3, Midjourney, Flux,
-SD3.5, Recraft, HiDream), in-the-wild social-media data (WildRF train), and a
-replay slice of CommunityForensics-Small, with web-realism augmentation
-(JPEG cascades, resize chains). GPT-4o and Ideogram were held out entirely.
+SD3.5, Recraft, HiDream), in-the-wild social media (WildRF train), and a
+replay slice of CommunityForensics-Small — plus, new in v3, non-photographic
+REAL classes that defeat most detectors: human paintings (WikiArt), hand-drawn
+anime, meme composites and webpage screenshots. GPT-4o and Ideogram were held
+out of training entirely.
 
 Preprocessing: shortest edge → 440 (bilinear) → center-crop 384 → [0,1] →
 ImageNet normalize. Input `pixel_values` 1×3×384×384 fp32, output `logit` 1×1.
 
 Held-out results through the extension pipeline @0.65 threshold:
-WildRF-test 97.6% balanced accuracy (mangled 97.1%), modern-generator eval
-98.3%. Full training + evaluation code: https://github.com/ashhart/Detectra
+WildRF-test 97.0% balanced accuracy (mangled 96.9%), modern-generator eval
+99.4% BA (TPR 98.9% incl. never-trained GPT-4o/Ideogram). Per-category TNR on
+difficult reals: paintings 99.8%, hand-drawn anime 100%, memes 97.6%, webpage
+screenshots 100%. Full training + evaluation code:
+https://github.com/ashhart/Detectra
 
-sha256(model.onnx) = 94fac18030826da368304e0cd79ff831df3ab6e46f95463a84b19793a9d2692d
+sha256(model.onnx) = 1414b9aafaa01a644ed706224973f09b53a1388282104409862df9893b1b962b
 
 ## Files
 - model.onnx — fp16 ONNX export (deployed by the extension)

model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94fac18030826da368304e0cd79ff831df3ab6e46f95463a84b19793a9d2692d
+oid sha256:1414b9aafaa01a644ed706224973f09b53a1388282104409862df9893b1b962b
 size 43684258

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bad21cbc9c941e41458695b981b8c9474f5a2a20ec185d1f8a15bf809d8384bf
+oid sha256:6ba0d02023ec9272f60c2ab61a53dae8dba815c38f809b11b86bf3aa15acdfab
 size 87262324