Upload README.md

README.md

@@ -1,4 +1,5 @@
 ---
+pretty_name: CommerceGen-Bench
 language:
 - zh
 license: cc-by-nc-nd-4.0
@@ -25,45 +26,93 @@ configs:
 
 # CommerceGen-Bench
 
-CommerceGen-Bench is an e-commerce product-poster generation/editing benchmark with 400 cases: 200 single-image cases and 200 two-image cases. This package contains test inputs and frozen instructions, not a training split or reference output images. Instructions and target selling-point text are in Chinese.
-
-This package contains 400 samples for generating or editing e-commerce product posters: 200 featuring a single image and 200 featuring two images. Each sample includes both fixed PE and no-PE prompts; input images are embedded directly into the Parquet file, eliminating the need for public image links or separate image folder uploads.
+CommerceGen-Bench is an e-commerce product-poster generation and editing benchmark containing 400 test cases: 200 single-image editing cases and 200 multi-image editing cases. Each case includes fixed with-PE and without-PE instructions, Chinese target selling-point text, embedded input images, and a three-level product taxonomy.
 
+The release contains test inputs only. It provides neither a training split nor reference output images. Input images are stored directly in the Parquet files, so no separate image directory or public image URL is required.
 
 ## Files and configurations
 
-- `data/single_image_test.parquet`: 200 single-image cases.
-- `data/multi_image_test.parquet`: 200 two-image cases.
-- `README.md`: dataset configuration, schema, protocol, and known limitations.
+- `data/single_image_test.parquet`: 200 single-image editing cases.
+- `data/multi_image_test.parquet`: 200 multi-image editing cases.
+- `README.md`: dataset card, schema, protocol, limitations, and license.
+
+The `single_image` and `multi_image` configurations each expose a `test` split containing 200 cases. Both prompt conditions share the same input images; there is no duplicated with-PE/without-PE image export.
 
-The `default` configuration loads all 400 cases. `single_image` and `multi_image` load their respective 200-case tracks. All configurations expose a `test` split. Both prompt conditions share the same image inputs; there is no duplicate PE/no-PE image export.
+There is intentionally no `default` configuration. Load the two tracks separately and concatenate them locally if a combined 400-case dataset is required. Keeping the physical Parquet files assigned to only one configuration avoids duplicated size accounting in the Hugging Face Dataset Viewer.
 
 ## Schema
 
 | Field | Type | Meaning |
 | --- | --- | --- |
-| `sample_id` | string | Stable anonymized identifier, `commercegen_0001` through `commercegen_0400`, preserving the original benchmark case order. |
+| `sample_id` | string | Stable anonymized identifier drawn from `commercegen_0001` through `commercegen_0420`. IDs are non-contiguous because 20 cases were removed during curation. |
 | `track` | string | `single_image_editing` or `multi_image_editing`. |
-| `product_image` | Image | First input: the product reference image. |
-| `logo_image` | Image or null | Second input: the auxiliary logo reference; null for single-image cases. |
-| `instruction_with_pe` | string | Frozen historical enhanced instruction, preserved from the actual PE run. |
-| `instruction_without_pe` | string | Frozen strict no-PE instruction: the five-requirement poster prefix, newline-separated selling points, and the strict suffix described below. |
-| `target_text_list` | list of strings | Selling-point strings to render in the generated poster. |
-| `category` | string | The existing source category label in Chinese; 15 labels are retained without inventing a new category hierarchy. |
-
-No `hard_case_tags`, unverified PE model name/version, training-isolation claims, business identifiers, source storage paths, credentials, or signed download URLs are added. Original physical source row order is recoverable from `sample_id`; row order after combining Parquet files should not be used as an identifier.
+| `product_image` | Image | Primary product input. It is the only image input for single-image cases and the first image input for multi-image cases. |
+| `logo_image` | Image or null | Auxiliary logo/reference input for multi-image cases; null for single-image cases. |
+| `instruction_with_pe` | string | Frozen enhanced instruction used in the historical with-PE evaluation. |
+| `instruction_without_pe` | string | Frozen strict no-PE instruction containing the five-requirement prefix, newline-separated selling points, and the strict suffix described below. |
+| `target_text_list` | list of strings | Selling-point strings expected to appear in the generated poster. |
+| `top_level_category` | string | One of 6 top-level product groups. |
+| `category` | string | One of 15 product categories. |
+| `subcategory` | string | Fine-grained product category. The current release contains 47 non-empty subcategories. |
+
+Business identifiers, credentials, signed download URLs, private storage locations, and source filesystem information are not included in this public release. Embedded image fields contain image bytes only and do not expose source file locations.
+
+## Dataset composition
+
+### Tracks
+
+| Track | Cases |
+| --- | ---: |
+| Single-image editing | 200 |
+| Multi-image editing | 200 |
+| **Total** | **400** |
+
+### Top-level categories
+
+| Top-level category | Cases | Share |
+| --- | ---: | ---: |
+| Beauty & Fashion | 157 | 39.25% |
+| Home & Decor | 114 | 28.50% |
+| 3C & Appliances | 50 | 12.50% |
+| Sports & Baby | 25 | 6.25% |
+| Food & Kitchen | 45 | 11.25% |
+| Others | 9 | 2.25% |
+| **Total** | **400** | **100.00%** |
+
+### Product categories
+
+| Category | Cases | Share |
+| --- | ---: | ---: |
+| Apparel | 57 | 14.25% |
+| Footwear | 41 | 10.25% |
+| Bags & Accessories | 28 | 7.00% |
+| Beauty & Care | 31 | 7.75% |
+| Home Textile | 22 | 5.50% |
+| Lighting | 20 | 5.00% |
+| Hardware | 40 | 10.00% |
+| Furniture | 32 | 8.00% |
+| Digital 3C | 27 | 6.75% |
+| Home Appliances | 23 | 5.75% |
+| Outdoor | 18 | 4.50% |
+| Baby & Toys | 7 | 1.75% |
+| Kitchenware | 24 | 6.00% |
+| Food & Beverage | 21 | 5.25% |
+| Pets & Miscellaneous | 9 | 2.25% |
+| **Total** | **400** | **100.00%** |
+
+The taxonomy was assigned according to the actual primary product shown in each case. Category counts are not artificially equalized.
 
 ## Image and task protocol
 
-Pass `product_image` first. For a multi-image case, pass `logo_image` second as a separate image; do not concatenate the inputs. Select exactly one instruction column for a given experiment and record that choice with the model settings.
+Pass `product_image` first. For a multi-image case, pass `logo_image` second as a separate image; do not concatenate the two inputs. Select exactly one instruction column for a given experiment and record the prompt condition together with all model settings.
 
-Images are losslessly encoded as RGB PNG bytes at their original dimensions, with source metadata removed. Their RGB pixels were checked against the source images and both historical PE/no-PE input snapshots. This export does not resize or pad images and does not prescribe a new output resolution. Model-specific resizing, output size, seed, and generation parameters should be recorded separately.
+The release does not prescribe a generation resolution. Model-specific input resizing, output size, sampling steps, guidance scale, random seed, and other generation parameters should be recorded separately.
 
-There is no confirmed curated ground-truth output in this package. Existing model-generated outputs are not labeled as ground truth and are intentionally excluded. This is an input benchmark, not a pixel-reference reconstruction dataset.
+Generated images should preserve the product identity and, for multi-image cases, accurately reproduce the auxiliary logo/reference condition. All requested selling-point text should be rendered correctly and completely according to the selected instruction.
 
 ## PE and no-PE protocol
 
-`instruction_with_pe` is the existing frozen `prompt.cn_long`; it is not regenerated. The historical PE generator identity/version has not been verified for this package and is not asserted here.
+`instruction_with_pe` contains the frozen enhanced instruction generated by Qwen3.8-Max for the experiments reported in the paper. These instructions are released as fixed benchmark inputs and are not regenerated at evaluation time.
 
 `instruction_without_pe` uses the historical strict C variant: the fixed five-requirement poster prefix, the unnumbered selling points joined with newlines, followed by exactly:
 
@@ -71,74 +120,123 @@ There is no confirmed curated ground-truth output in this package. Existing mode
 请严格按照我给的卖点内容进行生产,不允许添加新的卖点
 ```
 
-The historical B variant used by some open-source runs omits that final suffix. B and C are not identical. To reproduce a documented B run from this package, remove the final newline and that exact suffix only; do not otherwise rewrite the instruction. The packaged C instructions were checked against all 400 saved strict no-PE prompts.
+The ASCII comma above is preserved exactly as used in the historical evaluation prompt.
+
+The historical B variant used by some open-source runs omits the final suffix. B and C are not identical. To reproduce a documented B run, remove only the final newline and the exact suffix above; do not otherwise rewrite the instruction.
+
+Prompt line breaks are preserved. Comma-compacted prompt variants used in presentation slides are not substituted into the released instruction fields.
+
+## Evaluation
+
+The paper evaluates generated images using Gemini 3 Flash with five dimensions:
+
+- Product Fidelity (PF)
+- Instruction Following (IF)
+- Product Integration (PI)
+- Naturalness Detection (ND)
+- Artifacts Detection (AD)
+
+Overall is the arithmetic mean of the five dimension scores:
+
+```text
+Overall = (PF + IF + PI + ND + AD) / 5
+```
+
+Single-image and multi-image tracks must be reported separately. With-PE and without-PE conditions must also be reported separately.
+
+For reproducible comparisons, record the exact generation model version, inference configuration, random seed, judge model ID, judge decoding configuration, evaluation prompt version, number of repeated judgments, and score-aggregation procedure.
 
-Prompt line breaks are preserved. The comma-compacted English-template prompts used for presentation slides are not substituted into these data fields. This package does not make historical runs with different B/C instructions directly comparable by itself.
+Because commercial poster generation admits multiple valid layouts and visual treatments, this benchmark does not use a single pixel-level reference as the primary evaluation target.
 
-## Known limitations
+## Important reproducibility notes
 
 The historical PE instructions are preserved without silently repairing them:
 
-- 33 cases do not contain at least one exact target selling-point string after whitespace normalization (35 missing strings total). This is a substring check, not proof that the corresponding semantic meaning is absent.
-- 73 cases contain an image marker beyond the number of actual inputs, including references to `[image3]` although cases have at most two input images.
-- The historical PE instruction for `commercegen_0301` includes a merchant storefront address. It is preserved for run fidelity and needs review before public distribution.
+- 29 cases do not contain at least one exact target selling-point string after removing whitespace differences (31 missing strings in total). This substring check does not prove that the corresponding semantic meaning is absent.
+- 67 cases contain an image marker beyond the number of actual inputs, including references to `[image3]` although cases have at most two input images.
+- The historical PE instruction for `commercegen_0301` includes a merchant storefront address. This case must be reviewed before public distribution.
 
-Any corrected prompt set should be versioned separately rather than replacing the prompts used in historical evaluations. No claim that all selling-point text survives PE is made here.
+Any corrected prompt set should be released as a separately versioned configuration rather than silently replacing the prompts used in historical evaluations.
 
-## Load locally
+## Load from Hugging Face
 
-Use a compatible recent `datasets`, `pyarrow`, and `Pillow` environment:
+Install a recent compatible environment:
 
 ```bash
 python -m pip install -U datasets pyarrow Pillow
 ```
 
-Load both files after extracting the package:
+Load the two tracks separately:
 
 ```python
-from datasets import load_dataset
+from datasets import load_dataset, concatenate_datasets
 
-dataset = load_dataset(
-    "parquet",
-    data_files={"test": "CommerceGen-Bench/data/*_image_test.parquet"},
+single = load_dataset(
+    "TaobaoTmall-AlgorithmProducts/CommerceGen-Bench",
+    "single_image",
     split="test",
 )
+
+multi = load_dataset(
+    "TaobaoTmall-AlgorithmProducts/CommerceGen-Bench",
+    "multi_image",
+    split="test",
+)
+
+assert len(single) == 200
+assert len(multi) == 200
+
+dataset = concatenate_datasets([single, multi])
 assert len(dataset) == 400
 
-sample = dataset[0]
-images = [sample["product_image"]]
-if sample["logo_image"] is not None:
-    images.append(sample["logo_image"])
+sample = multi[0]
+input_images = [sample["product_image"], sample["logo_image"]]
 prompt = sample["instruction_with_pe"]
 ```
 
-After uploading the files to the intended Hugging Face dataset repository:
+Private repositories require authentication by an account with access.
+
+## Load local Parquet files
 
 ```python
-from datasets import load_dataset
+from datasets import load_dataset, concatenate_datasets
 
-dataset = load_dataset(
-    "TaobaoTmall-AlgorithmProducts/CommerceGen-Bench",
-    split="test",
-)
 single = load_dataset(
-    "TaobaoTmall-AlgorithmProducts/CommerceGen-Bench",
-    "single_image",
+    "parquet",
+    data_files={"test": "data/single_image_test.parquet"},
     split="test",
 )
+
 multi = load_dataset(
-    "TaobaoTmall-AlgorithmProducts/CommerceGen-Bench",
-    "multi_image",
+    "parquet",
+    data_files={"test": "data/multi_image_test.parquet"},
     split="test",
 )
+
+dataset = concatenate_datasets([single, multi])
+assert len(dataset) == 400
 ```
 
-Private repositories require authentication by an account with access. The repository identifier above is an intended destination, not a claim that it has already been created or uploaded.
+## License
+
+CommerceGen-Bench is released under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) License.
+
+- The dataset may be copied and redistributed for non-commercial purposes with appropriate attribution.
+- Commercial use is prohibited.
+- Modified or adapted versions of the dataset may not be publicly distributed.
 
-## Manual upload
+License terms: https://creativecommons.org/licenses/by-nc-nd/4.0/
 
-Create a private Hugging Face **dataset** repository named `CommerceGen-Bench`. Extract the ZIP, then upload `README.md` to the repository root and both Parquet files to `data/`. Do not upload only the ZIP: the Viewer needs the extracted Parquet files and configuration. A separate image directory or public image URL is unnecessary because the image bytes are embedded.
+Brand names and trademarks remain the property of their respective owners. Use of this dataset must also comply with applicable privacy, publicity, trademark, and other third-party rights.
 
-For ModelScope, the same Parquet files can be uploaded to a dataset repository, but ModelScope-specific Viewer/subset configuration and live loading have not been verified. Downloaded files can always be loaded locally with the Parquet example above. Keep private until rights are approved there as well.
+## Citation
+
+```bibtex
+@article{xu2026commercegenstudio,
+  title={CommerceGen-Studio: An End-to-End Framework for Commercial-Grade Main-Image Generation},
+  author={Xu, Xiaoli and Wu, Zhengtao and Xu, Zhengze and others},
+  year={2026}
+}
+```
 
-No repository has been created, and no files have been uploaded by this preparation step.
+Please replace this provisional citation with the full author list and arXiv or conference metadata when available.