Contamination MapVision / Model
FiLo, trained on VisA
- Organisation
- Institute of Automation, Chinese Academy of Sciences et al.
- Family
- FiLo · version filo_train_on_visa.pth + grounding_train_on_visa.pth
- Released
- 2024-04-21 (arXiv v1 submission date (https://arxiv.org/abs/2404.13671))
- Checkpoints
- FantasticGNU/FiLo filo_train_on_visa.pth, FantasticGNU/FiLo grounding_train_on_visa.pth
- Weights
- Open · Apache-2.0
- Links
- PaperCode
Training data disclosure: Full list
Level determined for 12% of benchmark cells.
- TestTest data seen1
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen0
- NoNot seen0
- ?Unknown7
“Similar to APRIL-GAN [4] and AnomalyCLIP [37], we conduct supervised training on the test set of one dataset and directly performed zero-shot testing on the other dataset.”
“Consistent with FiLo's experimental setup, we tested Grounding DINO fine-tuned on the VisA dataset on the MVTec dataset and tested Grounding DINO fine-tuned on the MVTec dataset on the VisA dataset.”
Built on
Everything these base models may have seen is inherited by FiLo, trained on VisA.
-
OpenAI CLIP
· disclosure: Partial list
Evidence
“We utilize the publicly available CLIP-L/14@336px model as our backbone, with frozen parameters for CLIP's Text Encoder and Image Encoder.”
arxiv.org · paper · Section 4.3 · retrieved 2026-10-15 “"--clip_pretrained", type=str, default="openai",”
raw.githubusercontent.com · code · train.py · retrieved 2026-10-15
Other declared training data
-
VisA · split: test · fine-tuning
Evidence
“Similar to APRIL-GAN [4] and AnomalyCLIP [37], we conduct supervised training on the test set of one dataset and directly performed zero-shot testing on the other dataset.”
arxiv.org · paper · Section 4.1 · retrieved 2026-10-15 -
VisA · split: unknown · fine-tuning
Split used to fine-tune Grounding DINO not stated.
Evidence
“Consistent with FiLo's experimental setup, we tested Grounding DINO fine-tuned on the VisA dataset on the MVTec dataset and tested Grounding DINO fine-tuned on the MVTec dataset on the VisA dataset.”
raw.githubusercontent.com · readme · Prepare Grounding DINO checkpoint · retrieved 2026-10-15 -
Pre-training data of base model grounding-dino (not tracked here) (broad web data: may contain benchmark test sets) · split: unknown
Evidence
“You can download our fine-tuned Grounding DINO model from the table below. We fine-tuned Grounding DINO using [MMDetection](https://github.com/open-mmlab/mmdetection).”
raw.githubusercontent.com · readme · Prepare Grounding DINO checkpoint · retrieved 2026-10-15
Benchmark cells
Industrial anomaly detection
Unknown MVTec AD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Unknown MVTec AD 2 · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Unknown MVTec LOCO AD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Test data seen VisA · in fine-tuning data: its test split was used for training
Evidence (1)
“Similar to APRIL-GAN [4] and AnomalyCLIP [37], we conduct supervised training on the test set of one dataset and directly performed zero-shot testing on the other dataset.”
Unknown Real-IAD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Unknown MPDD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Unknown BTAD · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Unknown KolektorSDD2 · unknown for its base model OpenAI CLIP: training data not disclosed in enough detail
Notes
Used for zero-shot results on MVTec AD.
Sources
- FiLo: Zero-Shot Anomaly Detection by Fine-Grained Description and High-Quality Localization (ACM MM 2024) · paper · retrieved 2026-10-15
- FiLo README (repo now CASIA-LMC-Lab/FiLo) · readme · retrieved 2026-10-15
- train.py · code · retrieved 2026-10-15