Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification"
Study scope and design Image-level prediction labels, source/evaluation manifests, 30 run records, statistical summaries, anonymous rust-assessment labels and retrospective integrity/sensitivity analyses for six maize disease/pest classification configurations with five seeds. Two seeds and optimiza...
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Conjunto de datos |
| Lenguaje: | inglés |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | Enlace permanente (DOI o Handle) |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1881700535056203776 |
|---|---|
| author | Hamilton Villamar-Barros Julián Coronel-Reyes Vanessa Vergara-Lozano Alexander Haro-Sarango |
| author_browse | Alexander Haro-Sarango Hamilton Villamar-Barros Julián Coronel-Reyes Vanessa Vergara-Lozano |
| author_facet | Hamilton Villamar-Barros Julián Coronel-Reyes Vanessa Vergara-Lozano Alexander Haro-Sarango |
| author_sort | Hamilton Villamar-Barros |
| collection | Producción científica |
| description | Study scope and design Image-level prediction labels, source/evaluation manifests, 30 run records, statistical summaries, anonymous rust-assessment labels and retrospective integrity/sensitivity analyses for six maize disease/pest classification configurations with five seeds. Two seeds and optimization checks were added after earlier external outcomes were known. Selection uses internal validation performance. The four evaluation axes differ in tasks/metrics. Native-pixel auditing detected residual transformed-image overlap across internal partitions. Exclusion sensitivity does not reconstruct clean training/selection, and candidate families do not establish biological independence. Source photographs and reviewer correspondence are excluded. Contents and file verification This record includes the data/code archive and all 30 selected model checkpoints listed in CHECKPOINT_FILES.csv, uploaded as separate files. All 31 files were verified after upload using file sizes and checksums. CC BY 4.0 applies to original derived data and documentation; original software and third-party components retain their applicable terms, as detailed in the archive's license and pretrained-model notices. Human-review scope Human-review status on 8 November 2026: all 190 critical pairs have recorded judgments, comprising 189 same-photograph or transformation judgments and one related capture; none are pending. Two contributors reviewed disjoint subsets of 171 and 19 pairs, rather than independently rating every pair; no inter-rater agreement is calculated. The four computationally exact critical RGB pairs are included in the 189 visual same-photograph judgments. All critical relationships were already covered by the conservative exclusion, leaving masks and scores unchanged. Coded reviewer IDs are used; private identities and free-text notes are excluded. The other 974 strong candidates were not visually adjudicated. Biological parents and acquisition metadata remain unconfirmed. Provenance and attribution Attribution notices distinguish pretrained-model terms from the proposed license for original derived data. Operational source labels include historical filename-based assignments; those labels do not uniformly establish collecting institution or country. The recovered 20 July source-audit closure records exclusion of pregenerated augmentation, without establishing complete biological independence. Licenses by component Original derived data and documentation: CC BY 4.0. Original software: MIT. ViT pretrained components: Apache-2.0; Swin pretrained components: MIT. MobileViT and torchvision components retain their applicable upstream terms. See the archive's pretrained-model notices and the corresponding repository notices. Source photographs, private reviewer identities and reviewer correspondence are excluded. |
| format | Dataset |
| id | openalex:W7225636726 |
| institution | UPeU |
| language | English |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| record_format | dspace |
| spelling | openalex:W7225636726Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification"Hamilton Villamar-BarrosJulián Coronel-ReyesVanessa Vergara-LozanoAlexander Haro-Sarangodomain shiftimage classificationmodel reproducibilitydata provenanceimage augmentationSmart Agriculture and AIDomain Adaptation and Few-Shot LearningAdvanced Neural Network ApplicationsStudy scope and design Image-level prediction labels, source/evaluation manifests, 30 run records, statistical summaries, anonymous rust-assessment labels and retrospective integrity/sensitivity analyses for six maize disease/pest classification configurations with five seeds. Two seeds and optimization checks were added after earlier external outcomes were known. Selection uses internal validation performance. The four evaluation axes differ in tasks/metrics. Native-pixel auditing detected residual transformed-image overlap across internal partitions. Exclusion sensitivity does not reconstruct clean training/selection, and candidate families do not establish biological independence. Source photographs and reviewer correspondence are excluded. Contents and file verification This record includes the data/code archive and all 30 selected model checkpoints listed in CHECKPOINT_FILES.csv, uploaded as separate files. All 31 files were verified after upload using file sizes and checksums. CC BY 4.0 applies to original derived data and documentation; original software and third-party components retain their applicable terms, as detailed in the archive's license and pretrained-model notices. Human-review scope Human-review status on 8 November 2026: all 190 critical pairs have recorded judgments, comprising 189 same-photograph or transformation judgments and one related capture; none are pending. Two contributors reviewed disjoint subsets of 171 and 19 pairs, rather than independently rating every pair; no inter-rater agreement is calculated. The four computationally exact critical RGB pairs are included in the 189 visual same-photograph judgments. All critical relationships were already covered by the conservative exclusion, leaving masks and scores unchanged. Coded reviewer IDs are used; private identities and free-text notes are excluded. The other 974 strong candidates were not visually adjudicated. Biological parents and acquisition metadata remain unconfirmed. Provenance and attribution Attribution notices distinguish pretrained-model terms from the proposed license for original derived data. Operational source labels include historical filename-based assignments; those labels do not uniformly establish collecting institution or country. The recovered 20 July source-audit closure records exclusion of pregenerated augmentation, without establishing complete biological independence. Licenses by component Original derived data and documentation: CC BY 4.0. Original software: MIT. ViT pretrained components: Apache-2.0; Swin pretrained components: MIT. MobileViT and torchvision components retain their applicable upstream terms. See the archive's pretrained-model notices and the corresponding repository notices. Source photographs, private reviewer identities and reviewer correspondence are excluded.Datasethttps://doi.org/10.5281/zenodo.24064538W7225636726enZenodo (CERN European Organization for Nuclear Research)2026-11-09 |
| spellingShingle | domain shift image classification model reproducibility data provenance image augmentation Smart Agriculture and AI Domain Adaptation and Few-Shot Learning Advanced Neural Network Applications Hamilton Villamar-Barros Julián Coronel-Reyes Vanessa Vergara-Lozano Alexander Haro-Sarango Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification" |
| title | Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification" |
| title_full | Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification" |
| title_fullStr | Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification" |
| title_full_unstemmed | Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification" |
| title_short | Research data and reproducibility records for "Class–Source Confounding and Architecture-Selection Instability under Domain Shift: A Multi-Source Benchmark for Maize Disease and Pest Image Classification" |
| title_sort | research data and reproducibility records for class source confounding and architecture selection instability under domain shift a multi source benchmark for maize disease and pest image classification |
| topic | domain shift image classification model reproducibility data provenance image augmentation Smart Agriculture and AI Domain Adaptation and Few-Shot Learning Advanced Neural Network Applications |
| topic_facet | domain shift image classification model reproducibility data provenance image augmentation Smart Agriculture and AI Domain Adaptation and Few-Shot Learning Advanced Neural Network Applications |
| url | https://doi.org/10.5281/zenodo.24064538 |
| work_keys_str_mv | AT hamiltonvillamarbarros researchdataandreproducibilityrecordsforclasssourceconfoundingandarchitectureselectioninstabilityunderdomainshiftamultisourcebenchmarkformaizediseaseandpestimageclassification AT juliancoronelreyes researchdataandreproducibilityrecordsforclasssourceconfoundingandarchitectureselectioninstabilityunderdomainshiftamultisourcebenchmarkformaizediseaseandpestimageclassification AT vanessavergaralozano researchdataandreproducibilityrecordsforclasssourceconfoundingandarchitectureselectioninstabilityunderdomainshiftamultisourcebenchmarkformaizediseaseandpestimageclassification AT alexanderharosarango researchdataandreproducibilityrecordsforclasssourceconfoundingandarchitectureselectioninstabilityunderdomainshiftamultisourcebenchmarkformaizediseaseandpestimageclassification |