saving image in memory and sending via api

api/app.py

@@ -1,6 +1,10 @@
-from flask import Flask, render_template, send_from_directory
+from flask import Flask, render_template, send_from_directory,  send_file
 import subprocess
 import os
+import io
+from model.src.utils.arg_parser import eval_parse_args  # Nuovo import corretto
+
+from model.src import eval
 
 app = Flask(__name__)
 
@@ -14,7 +18,8 @@ def index():
 
 @app.route('/generate-design', methods=['POST'])
 def generate_design():
-    # Esegui il comando di generazione
+    
+    '''# Esegui il comando di generazione
     command = [
         'python', '/api/model/src/eval.py',
         '--dataset_path', '/api/model/assets/data/vitonhd',
@@ -30,13 +35,37 @@ def generate_design():
     ]
     
     # Esegui il comando senza gestire il JSON, lascia che l'errore venga stampato
-    subprocess.run(command, check=True)
+    subprocess.run(command, check=True)'''
+    try:
+        # Creiamo una lista di argomenti come quelli che passeresti via CLI
+        args_list = [
+            '--dataset_path', '/api/model/assets/data/vitonhd',
+            '--batch_size', '1',
+            '--mixed_precision', 'fp16',
+            '--output_dir', '/api/output',  # Se serve ancora questo argomento
+            '--save_name', 'generato_paired',
+            '--num_workers_test', '4',
+            '--sketch_cond_rate', '0.2',
+            '--dataset', 'vitonhd',
+            '--start_cond_rate', '0.0',
+            '--test_order', 'paired'
+        ]
+        
+        # Simuliamo l'arg_parser con la lista di argomenti
+        args = eval_parse_args(args_list)
 
-    files = os.listdir(output_dir)
-    for file in files:
-            print(file)
-    
-    return '', 204  # Nessun contenuto
+         # Esegui la funzione `main()` di eval.py passando gli argomenti
+        final_image = eval.main(args)
+
+        # Converti l'immagine in un buffer di memoria
+        img_io = io.BytesIO()
+        final_image.save(img_io, 'JPEG')  # Cambia il formato se necessario
+        img_io.seek(0)  # Riporta il puntatore all'inizio del buffer
+
+        # Invia l'immagine come file scaricabile
+        return send_file(img_io, mimetype='image/jpeg', as_attachment=True, attachment_filename='generated_image.jpg')
+    except Exception as e:
+        return str(e), 500
 
 # Route per servire l'immagine generata
 @app.route('/api/output/generato_paired_paired/images/<filename>')

api/model/src/eval.py

@@ -19,6 +19,7 @@ from mgd_pipelines.mgd_pipe_disentangled import MGDPipeDisentangled
 from utils.arg_parser import eval_parse_args
 from utils.image_from_pipe import generate_images_from_mgd_pipe
 from utils.set_seeds import set_seed
+from PIL import Image
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
 check_min_version("0.10.0.dev0")
@@ -133,7 +134,7 @@ def main() -> None:
 
         val_pipe.enable_attention_slicing()
         test_dataloader = accelerator.prepare(test_dataloader)
-        generate_images_from_mgd_pipe(
+        final_image = generate_images_from_mgd_pipe(
             test_order=args.test_order,
             pipe=val_pipe,
             test_dataloader=test_dataloader,
@@ -149,7 +150,7 @@ def main() -> None:
             disentagle=False,
             seed=args.seed,
         )
-
+        return final_image  # Now returning the generated image
 
 if __name__ == "__main__":
     main()

api/model/src/utils/image_from_pipe.py

@@ -6,6 +6,7 @@ import torchvision.transforms as T
 from diffusers.pipeline_utils import DiffusionPipeline
 from torch.utils.data import DataLoader
 from src.utils.image_composition import compose_img, compose_img_dresscode
+from PIL import Image
 
 
 @torch.inference_mode()
@@ -46,12 +47,12 @@ def generate_images_from_mgd_pipe(
         Returns:
         None
     """ 
-    assert(save_name != ""), "save_name must be specified"
-    assert(output_dir != ""), "output_dir must be specified"
+    #assert(save_name != ""), "save_name must be specified"
+    #assert(output_dir != ""), "output_dir must be specified"
 
-    path = os.path.join(output_dir, f"{save_name}_{test_order}", "images")
+    #path = os.path.join(output_dir, f"{save_name}_{test_order}", "images")
 
-    os.makedirs(path, exist_ok=True)
+    #os.makedirs(path, exist_ok=True)
     generator = torch.Generator("cuda").manual_seed(seed)
 
     for batch in tqdm(test_dataloader):
@@ -122,6 +123,7 @@ def generate_images_from_mgd_pipe(
                         im_head = face
                         )
                 
-                final_img = T.functional.to_pil_image(final_img)
-                final_img.save(
-                    os.path.join(path, batch["im_name"][i].replace(".jpg", ext)))
+                return T.functional.to_pil_image(final_img)
+                #final_img = T.functional.to_pil_image(final_img)
+                #final_img.save(
+                #    os.path.join(path, batch["im_name"][i].replace(".jpg", ext)))