Recent studyBlind TTS Elo is live. Compare two anonymous voice samples, vote after listening, and help separate real preference signal from noise.Vote in the study ->
Codesota · Computer Vision · Document Image Classification · tobacco-3482Tasks/Computer Vision/Document Image Classification
Document Image Classification · benchmark dataset · 2020 · EN

tobacco-3482.

Dataset from Papers With Code

Submit a result ↵
§ 01 · Leaderboard

Best published scores.

18 results indexed across 3 metrics. Shaded row marks current SOTA; ties broken by submission date.

Primary
accuracy · higher is better
All metrics
accuracy, memory, training-time-hours
accuracy· primary
14 rows
#ModelOrgSubmittedPaper / codeaccuracy
01HEADoC-Large—Oct 2025HEADoC: Highly Efficient and Accurate Document Classifier Optimized Using Semantic Distances96.66
02HEADoC-Base—Oct 2025HEADoC: Highly Efficient and Accurate Document Classifier Optimized Using Semantic Distances95.98
03DocXClassifier-FPNSaifullah et al.Jun 2024DocXclassifier: towards a robust and interpretable deep neural network for document image classification95.71
04DocXClassifier-L—Mar 2022papers-with-code · code95.57
05Multimodal Side-Tuning (MobileNetV2)—Jan 2023Multimodal Side-Tuning for Document Classification · code90.50
06Multimodal Side-Tuning (ResNet50)—Jan 2023Multimodal Side-Tuning for Document Classification · code90.30
07DiT-BaseMicrosoftDec 2024Label Errors in the Tobacco3482 Dataset84.10
08DocBert [DOCBERT]—Jun 2021Efficient Document Image Classification Using Region-Bas…82.30
09BERT [BERT]—Jun 2021Efficient Document Image Classification Using Region-Bas…79
10Eff-GNN + Word2Vec [word2vec] + Image Embedding—Jun 2021Efficient Document Image Classification Using Region-Bas…77.50
11Eff-GNN + Word2Vec [word2vec]—Jun 2021Efficient Document Image Classification Using Region-Bas…73.50
12Optimized Text CNN—Apr 2020Light-Weighted CNN for Text Classification · code46
13Lightweight TextCNN with Dual Optimizer—Apr 2020Light-Weighted CNN for Text Classification · code43.50
14Lightweight Text CNN—Apr 2020Light-Weighted CNN for Text Classification · code42
memory
1 row
#ModelOrgSubmittedPaper / codememory
01VGG—Jun 2021Efficient Document Image Classification Using Region-Bas…7.08
training-time-hours
3 rows
#ModelOrgSubmittedPaper / codetraining-time-hours
01Optimized Text CNN—Apr 2020Light-Weighted CNN for Text Classification · code2.00
02Lightweight Text CNN—Apr 2020Light-Weighted CNN for Text Classification · code1.00
03Lightweight TextCNN with Dual Optimizer—Apr 2020Light-Weighted CNN for Text Classification · code0.430
Fig 2 · Rows sorted by score within each metric. Shaded row marks SOTA. Dates reflect model or paper release where available, otherwise the date Codesota accessed the source.
§ 03 · Progress

5 steps
of state of the art.

Each row below marks a model that broke the previous record on accuracy. Intermediate submissions are kept in the leaderboard above; only SOTA-setting entries are re-listed here.

Higher scores win. Each subsequent entry improved upon the previous best.

SOTA line · accuracy
  1. Apr 16, 2020Optimized Text CNN46
  2. Jun 25, 2021DocBert [DOCBERT]82.30
  3. Mar 17, 2022DocXClassifier-L95.57
  4. Jun 25, 2024DocXClassifier-FPNSaifullah et al.95.71
  5. Oct 12, 2025HEADoC-Large96.66
Fig 3 · SOTA-setting models only. 5 entries span Apr 2020 → Oct 2025.
§ 04 · Literature

3 papers
tied to this benchmark.

Every paper below corresponds to at least one row in the leaderboard above. Click through for the arXiv preprint and, when available, the reference implementation.

§ 06 · Contribute

Have a score that beats
this table?

Submit a checkpoint and a reproduction script. We will run it, publish the score, and — if it takes the top — annotate the step on the progress chart with your name.

Submit a result ↵Read submission guide
What a submission needs
  • 01A public checkpoint or API endpoint
  • 02A reproduction script with frozen commit + seed
  • 03Declared evaluation environment (Python, deps)
  • 04One row per metric declared by this dataset
  • 05A contact so we can follow up on discrepancies