remove unused params, tune SAM2 params

app.py

@@ -9,8 +9,7 @@ from PIL import Image
 import dounseen.utils as dounseen_utils
 import cv2
 
-from utils.models import load_sam2_models, CHECKPOINT_NAMES, MODE_NAMES, \
-    MASK_GENERATION_MODE, BOX_PROMPT_MODE, load_dounseen_model
+from utils.models import load_sam2_models, CHECKPOINT_NAMES, load_dounseen_model
 
 # TODO add presentation on YouTube and add link here
 MARKDOWN = """
@@ -89,9 +88,10 @@ with gr.Blocks() as demo:
             with gr.Row():
                 object_image_1 = gr.Image(type='pil', label=f'Object Image 1')
                 object_image_2 = gr.Image(type='pil', label=f'Object Image 2')
-                object_image_3 = gr.Image(type='pil', label=f'Object Image 3')
             with gr.Row():
+                object_image_3 = gr.Image(type='pil', label=f'Object Image 3')
                 object_image_4 = gr.Image(type='pil', label=f'Object Image 4')
+            with gr.Row():
                 object_image_5 = gr.Image(type='pil', label=f'Object Image 5')
                 object_image_6 = gr.Image(type='pil', label=f'Object Image 6')
             object_images = [object_image_1, object_image_2, object_image_3, object_image_4, object_image_5, object_image_6]

utils/models.py

@@ -6,10 +6,6 @@ from sam2.build_sam import build_sam2
 from sam2.sam2_image_predictor import SAM2ImagePredictor
 from dounseen.core import UnseenClassifier
 
-BOX_PROMPT_MODE = "box prompt"
-MASK_GENERATION_MODE = "mask generation"
-MODE_NAMES = [BOX_PROMPT_MODE, MASK_GENERATION_MODE]
-
 CHECKPOINT_NAMES = ["tiny", "small", "base_plus", "large"]
 CHECKPOINTS = {
     "tiny": ["sam2_hiera_t.yaml", "models/sam2/sam2_hiera_tiny.pt"],
@@ -27,12 +23,12 @@ def load_sam2_models(
         model = build_sam2(config, checkpoint, device=device)
         mask_generators[key] = SAM2AutomaticMaskGenerator(
             model=model,
-            points_per_side=16,
-            points_per_batch=64,
+            points_per_side=64,
+            points_per_batch=128,
             pred_iou_thresh=0.7,
             stability_score_thresh=0.92,
             stability_score_offset=0.7,
-            crop_n_layers=0,
+            crop_n_layers=1,
             box_nms_thresh=0.7,
         )
     return mask_generators