update app.py

app.py

@@ -434,7 +434,9 @@ def concatenate_images_vertical(image1, image2):
 
     return new_image
 
-def relate_anything(input_image, k):
+def relate_anything(input_image_mask, k):
+    logger.info(f'relate_anything_1_')
+    input_image = input_image_mask['image']
     w, h = input_image.size
     max_edge = 1500
     if w > max_edge or h > max_edge:
@@ -442,12 +444,14 @@ def relate_anything(input_image, k):
         new_size = (int(w / ratio), int(h / ratio))
         input_image.thumbnail(new_size)
     
+    logger.info(f'relate_anything_2_')
     # load image
     pil_image = input_image.convert('RGBA')
     image = np.array(input_image)
     sam_masks = sam_mask_generator.generate(image)
     filtered_masks = sort_and_deduplicate(sam_masks)
 
+    logger.info(f'relate_anything_3_')
     feat_list = []
     for fm in filtered_masks:
         feat = torch.Tensor(fm['feat']).unsqueeze(0).unsqueeze(0).to(device)
@@ -455,6 +459,7 @@ def relate_anything(input_image, k):
     feat = torch.cat(feat_list, dim=1).to(device)
     matrix_output, rel_triplets = ram_model.predict(feat)
 
+    logger.info(f'relate_anything_4_')
     pil_image_list = []
     for i, rel in enumerate(rel_triplets[:k]):
         s,o,r = int(rel[0]),int(rel[1]),int(rel[2])
@@ -473,6 +478,7 @@ def relate_anything(input_image, k):
         concate_pil_image = concatenate_images_vertical(current_pil_image, title_image)
         pil_image_list.append(concate_pil_image)
 
+    logger.info(f'relate_anything_5_')
     yield pil_image_list