--- language: - en license: mit task_categories: - text-to-image - unconditional-image-generation pretty_name: Housey Home v2 dataset_info: features: - name: image dtype: image - name: prompt dtype: string - name: hash dtype: string - name: generated_by dtype: string - name: quantity dtype: string - name: plain_background dtype: string - name: has_neighbors dtype: string - name: style sequence: string - name: size sequence: sequence: string splits: - name: train num_bytes: 873060466.6 num_examples: 3032 download_size: 872503009 dataset_size: 873060466.6 configs: - config_name: default data_files: - split: train path: data/train-* tags: - art --- # Housey Home v2 - Like v1 never happened I was in the process of producing a fully synthetic dataset for ungrounded image generation using an unconventional combination of layers. As such, I needed a dataset of highly similar objects with 'themes'. In order to produce `log(x, y)` combinations of options in the final model. This is that dataset. The initial ( 07/15/2024 ) release includes ~2k unique houses, each processed using a VQA, [ybelkada/blip-vqa-base](https://huggingface.co/Salesforce/blip-vqa-base), to be precise. Release 2 includes ~3k unique houses, each processed more efficiently and using a more powerful GPU via parallel model loading followed by sequential but accelerated inference and diffusion with each. ``` # This code procedurally generates a simple description based upon the input lists of string selections. Feel free to expand or use as MIT. word_salad = [ ["cartoon", "happy", "goofy", "virant", "whymsical"], ["realistic", "ultra hd", "best quality", "high quality", "masterpiece"], ["horrifying", "nightmare", "wicked", "evil", "dark", "creepy", "scary"] ] sizes = [ ["large mansion", "mansion", "manor", "estate", "enournmous house", "palace"], ["medium house", "home", "residence", "big house", "large house"], ["tiny shack", "cottage", "small house", "miniature house", "dinimutive house", "shack"] ] post_prompt = ", ".join(chosen_salad := random.choice(word_salad)) + ", " + ", ".join(chosen_sizes := random.choice(sizes)) description = "A " + chosen_sizes[0].split(" ")[0] + ", " + chosen_salad[0] + " " + chosen_sizes[0].split(" ")[1] ```