AnomalyThink-Qwen3-VL-8B-KCR

AnomalyThink-Qwen3-VL-8B-KCR

Qwen3-VL-8B trained from base on a Keep-Correct-Revise corpus built from its own GRPO rollouts. DS-MVTec 84.39 / VisA 76.63, the best VisA score of any model in the project.

Post-submission research artefact. The thesis reports Qwen3-VL-8B only as a corpus-transfer check (base 78.68 / 64.45, trained on the Qwen2.5-VL KCR corpus 85.82 / 76.45, Table 6.13). The three Qwen3-VL-8B models in this family (AnomalyThink-Qwen3-VL-8B-SFT, AnomalyThink-Qwen3-VL-8B-SFT-GRPO, AnomalyThink-Qwen3-VL-8B-KCR) were trained after the thesis was submitted, in September 2026, by running the thesis recipe end to end on this backbone. They are not thesis results and are released for reproducibility.

Results (MMAD subsets, balanced accuracy, strict scoring)

Evaluation protocol. One harness for every row: the DS-MVTec (1,670 images) and VisA (2,141 images) subsets of MMAD, one image per prompt, the training prompt with the one-line system message "Please answer by yes or no", greedy decoding, at most 1,024 new tokens, images capped at 262,144 pixels, vLLM generation. Balanced accuracy is the mean of sensitivity and specificity. Strict scoring: a generation with no parsable <answer> counts as wrong. Per-sample evaluation files are next to the weights (eval_*.json).

The pattern of the thesis reproduces on this backbone: GRPO carries DS-MVTec, and the corrected corpus built from the GRPO policy's rollouts carries VisA, where 76.63 is the highest VisA score of any model trained in the project.

Training

• Base: Qwen/Qwen3-VL-8B-Instruct, trained from the base weights, not continued from the GRPO policy.
• Corpus: qwen3_kcr/sft_qwen3_C_train.json of aacudad/AnomalyThink, 5,998 traces balanced by verdict (2,999 anomalous). Built by rolling AnomalyThink-Qwen3-VL-8B-SFT-GRPO out eight times at temperature 0.7 over the 10,236 Real-IAD training images, scoring every rollout with the local rewards and a Gemini-3-Flash faithfulness judge, and routing items to keep (8,988), correct (1,248) or rewrite (2,464) at pool level, then assembling the corpus on the 6,000 SFT images only.
• Recipe: identical to the SFT model, vision encoder frozen, learning rate 1e-5, cosine, effective batch 32, 4 epochs. This is epoch 2 (step 376), the DS-MVTec-best epoch; the four epochs scored 82.67 / 71.40, 84.39 / 76.63, 84.14 / 71.96 and 83.67 / 70.32.

What Keep-Correct-Revise (KCR) means

An SFT plus GRPO policy is sampled eight times on every training image. A trace is kept when the verdict is right and a Gemini-3-Flash judge finds the reasoning grounded, corrected by the teacher when every rollout was wrong, and revised by the teacher when the verdict was right but the reasoning was weakly grounded. A fresh model is then fine-tuned from the base weights on the curated corpus, so the final model is supervised and the reinforcement-learned policy is the source of its training data.

Usage

The checkpoint was saved with transformers 5.0.0 and loads directly with transformers>=5.0:

from transformers import AutoProcessor, AutoModelForImageTextToText
import torch
repo = "aacudad/AnomalyThink-Qwen3-VL-8B-KCR"
processor = AutoProcessor.from_pretrained(repo)
model = AutoModelForImageTextToText.from_pretrained(repo, torch_dtype=torch.bfloat16, device_map="auto")
msgs = [{"role": "system", "content": "Please answer by yes or no"},
        {"role": "user", "content": [{"type": "image", "image": "part.jpg"},
         {"type": "text", "text": "Analyze the provided image of the pcb. Determine if there are any anomalies present. If an anomaly is detected, specify its type and location, and provide a detailed reasoning for your conclusion."}]}]
inputs = processor.apply_chat_template(msgs, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=1024, do_sample=False)
print(processor.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

The output is <think>...</think> followed, for anomalies, by <location>, <type> and <answer>Yes</answer>, or <answer>No</answer> for a normal part. For transformers 4.57 or vLLM 0.10.x use the files in transformers4_vllm/ (the stock Qwen3-VL-8B-Instruct tokenizer and processor files and a config with rope_theta written into text_config) next to model.safetensors; this is how every number above was produced.

Intended use and limitations

Research artefact for explainable industrial anomaly detection on top-down product images. Single seed, no human evaluation of the explanations, selection on DS-MVTec with VisA reported at the same checkpoint. The public benchmark images may have been seen during backbone pretraining, which cannot be ruled out for any model in this comparison.

Citation

@mastersthesis{acudad2026reasoning,
  author  = {Acudad, A.},
  title   = {Reasoning-Enhanced Vision-Language Models for Explainable Industrial Anomaly Detection},
  school  = {Delft University of Technology},
  year    = {2026},
  type    = {Master's thesis},
  url     = {https://resolver.tudelft.nl/uuid:65c62420-79c0-447f-b095-7fb11d4474fc}
}

Thesis: https://resolver.tudelft.nl/uuid:65c62420-79c0-447f-b095-7fb11d4474fc. Code and evaluation files: https://github.com/aacudad/IAD-VLMs.

License

Apache-2.0, derived from Qwen/Qwen3-VL-8B-Instruct.