Evaluation

Directly run scripts/eval/run_geneval.sh to evaluate GenEVAL. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• we Set --think for native prompt enhancing.
• See GenEval for original GenEval prompts.

DPGBench

Evaluation

Directly run scripts/eval/run_dpgbench.sh to evaluate DPGBench. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• See DPGBench for original DPGBench prompts.

WISE

We modify the code in WISE for faster evaluation and download prompt json files.

Evaluation

Directly run scripts/eval/run_wise.sh to evaluate WISE. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• Use think for World Knowledge-Enhanced Textual Reasoning.
• scripts/eval/run_wise_refine.sh support both World Knowledge-Enhanced Textual Reasoning and Fine-grained Editing-like Visual Refinement through twice thinking and correct initial image.

GEdit-Bench

We adopt the code in GEdit-Bench for evaluation.

Evaluation

Modify the model path, the output path in scripts/eval/run_gedit.sh. Then, run the following command:

bash script/eval/run_gedit.sh

KRIS

We modify the code in KRIS-Bench for faster evaluation.

Data prepration

Please download the benchmark data from KRIS-Bench and and place it in the KRIS_Bench directory.

Evaluation

Directly run scripts/eval/run_kris.sh to evaluate KRIS-Bench. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• Use think for World Knowledge-Enhanced Textual Reasoning.
• scripts/eval/run_kris_refine.sh support both World Knowledge-Enhanced Textual Reasoning and Fine-grained Editing-like Visual Refinement through twice thinking and correct initial image.

ImgEdit

We modify the code in ImgEdit for faster evaluation.

Data prepration

Please download the benchmark data from ImgEdit-Bench and and place it in the Benchmark directory.

Evaluation

Directly run scripts/eval/run_imgedit.sh to evaluate ImgEdit-Bench. The output will be saved in $output_path.

UniGenBench

We create the eval code of UniGenBench for faster evaluation.

Data prepration

Please download the benchmark data from UniGenBench and place it in the Benchmark directory.

Evaluation

Directly run scripts/eval/run_unigenbench.sh to evaluate UniGenBench. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• scripts/eval/run_unigenbench_refine.sh support Fine-grained Editing-like Visual Refinement improve initial image quality from object presence, attribute accuracy, style consistency, and aesthetic quality

T2I-CoreBench

We create the eval code of T2I-CoreBench for faster evaluation.

Data prepration

Please download the benchmark data from T2I-CoreBench and and place it in the Benchmark directory.

Evaluation

Directly run scripts/eval/run_corebench.sh to evaluate T2I-CoreBench. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• scripts/eval/run_corebench_refine.sh support Fine-grained Editing-like Visual Refinement improve initial image quality from object presence, attribute accuracy, style consistency, and aesthetic quality

UniREditBench

We merge the eval code of UniREditBench for faster evaluation.

Data prepration

Please download the benchmark data from UniREditBench and and place it in the Benchmark directory.

Evaluation

Directly run scripts/eval/run_unireditbench.sh to evaluate UniREditBench. The output will be saved in $output_path.

• Set $model_path and $output_path for the path for checkpoint and log.
• Use think for World Knowledge-Enhanced Textual Reasoning.