switch to stable diffusion 2 for generation, euler scheduler, add xformers

app.py

@@ -18,19 +18,19 @@ from saac.prompt_generation.prompts import generate_prompts,generate_occupations
 from saac.prompt_generation.prompt_utils import score_prompt
 from saac.image_analysis.process import process_image_pil
 from saac.evaluation.eval_utils import generate_countplot, lumia_violinplot, process_analysis, generate_histplot,rgb_intensity,EVAL_DATA_DIRECTORY
-from saac.evaluation.evaluate import evaluate_gender_by_adjectives,evaluate_gender_by_occupation,evaluate_skin_by_adjectives,evaluate_skin_by_occupation
+from saac.evaluation.evaluate import evaluate_by_adjectives,evaluate_by_occupation
 from datasets import load_dataset
-from diffusers import DiffusionPipeline, PNDMScheduler
+from diffusers import DiffusionPipeline, PNDMScheduler,EulerDiscreteScheduler
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 STABLE_MODELS = ["Stable Diffusion v1.5", "Midjourney"]
 results = dict()
-results[STABLE_MODELS[0]] = process_analysis(os.path.join(EVAL_DATA_DIRECTORY,'raw',"stable_diffusion_raw_processed.csv"))
-results[STABLE_MODELS[1]] = process_analysis(os.path.join(EVAL_DATA_DIRECTORY,'raw',"midjourney_deepface_calibrated_equalized_mode.csv"))
+results[STABLE_MODELS[0]] = process_analysis(os.path.join(EVAL_DATA_DIRECTORY,'raw',"stable_diffusion_raw_processed.csv"),filtered=True)
+results[STABLE_MODELS[1]] = process_analysis(os.path.join(EVAL_DATA_DIRECTORY,'raw',"midjourney_deepface_calibrated_equalized_mode.csv"),filtered=True)
 
-scheduler = PNDMScheduler.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="scheduler", prediction_type="v_prediction",revision="fp16",
-                    torch_dtype=torch.float16)
-pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", scheduler=scheduler)
+# scheduler = PNDMScheduler.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="scheduler", prediction_type="v_prediction",revision="fp16",torch_dtype=torch.float16)
+esched = EulerDiscreteScheduler.from_pretrained("stabilityai/stable-diffusion-2-base",subfolder="scheduler")
+pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-base", scheduler=esched,)
 pipe = pipe.to(device)
 
 LOOKS = sorted(list(generate_traits()['tag']))#["beautiful", "stunning", "handsome", "ugly", "plain", "repulsive", "arrogant", "trustworthy"]
@@ -48,8 +48,7 @@ def fig2img(fig):
 
 def trait_graph(model,hist=True):
     tda_res,occ_res = results[model]
-    pass_gen = evaluate_gender_by_adjectives(tda_res)
-    pass_skin = evaluate_skin_by_adjectives(tda_res)
+    pass_gen,pass_skin = evaluate_by_adjectives(adjective_df=tda_res)
     fig = None
     if not hist:
         fig = generate_countplot(tda_res, 'tda_sentiment_val', 'gender_detected_val',
@@ -78,10 +77,9 @@ def trait_graph(model,hist=True):
     return pass_skin,pass_gen,fig2img(fig2),fig2img(fig)
 def occ_graph(model):
     tda_res,occ_result = results[model]
-    pass_skin = evaluate_skin_by_occupation(occ_result)
-    pass_gen = evaluate_gender_by_occupation(occ_result)
+    pass_gen,pass_skin = evaluate_by_occupation(occupation_df=occ_result)
     fig = generate_histplot(occ_result, 'a_median', 'gender_detected_val',
-                   title='Gender Distribution by Median Annual Salary',
+                   title='Gender Distribution by Median Annualg Salary',
                    xlabel= 'Median Annual Salary',
                    ylabel= 'Count',)
     fig2 = lumia_violinplot(df=occ_result, x_col='a_median',

requirements.txt

@@ -3,4 +3,5 @@ gradio
 transformers
 diffusers
 torch
-accelerate
+accelerate
+xformers