enable onnx inference

Y-T-G · Aug 11, 2023 · d685958 · d685958
1 parent d8305f8
commit d685958
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 17 deletions.
diff --git a/app.py b/app.py
@@ -56,7 +56,7 @@ def get_prompt(click_state, click_input):
         "prompt_type": ["click"],
         "input_point": click_state[0],
         "input_label": click_state[1],
-        "multimask_output": "True",
+        "multimask_output": "False",
     }
     return prompt
 

diff --git a/utils/base_segmenter.py b/utils/base_segmenter.py
@@ -24,8 +24,10 @@ def __init__(self, sam_pt_checkpoint, sam_onnx_checkpoint, model_type, device="c
             from mobile_sam import sam_model_registry, SamPredictor
             from onnxruntime import InferenceSession
             self.ort_session = InferenceSession(sam_onnx_checkpoint)
+            self.predict = self.predict_onnx
         else:
             from segment_anything import sam_model_registry, SamPredictor
+            self.predict = self.predict_pt
 
         self.model = sam_model_registry[model_type](checkpoint=sam_pt_checkpoint)
         self.model.to(device=self.device)
@@ -51,7 +53,7 @@ def reset_image(self):
         self.predictor.reset_image()
         self.embedded = False
 
-    def predict(self, prompts, mode, multimask=True):
+    def predict_pt(self, prompts, mode, multimask=True):
         """
         image: numpy array, h, w, 3
         prompts: dictionary, 3 keys: 'point_coords', 'point_labels', 'mask_input'
@@ -115,17 +117,20 @@ def predict_onnx(self, prompts, mode, multimask=True):
                 "orig_im_size": prompts["orig_im_size"],
             }
             masks, scores, logits = self.ort_session.run(None, ort_inputs)
+            masks = masks > self.predictor.model.mask_threshold
 
         elif mode == "mask":
             ort_inputs = {
                 "image_embeddings": self.image_embedding,
-                "point_coords": prompts["point_coords"],
+                "point_coords": np.zeros((len(prompts["point_labels"]), 2), dtype=np.float32),
                 "point_labels": prompts["point_labels"],
                 "mask_input": prompts["mask_input"],
                 "has_mask_input": np.ones(1, dtype=np.float32),
                 "orig_im_size": prompts["orig_im_size"],
             }
             masks, scores, logits = self.ort_session.run(None, ort_inputs)
+            masks = masks > self.predictor.model.mask_threshold
+
         elif mode == "both":  # both
             ort_inputs = {
                 "image_embeddings": self.image_embedding,
@@ -136,7 +141,9 @@ def predict_onnx(self, prompts, mode, multimask=True):
                 "orig_im_size": prompts["orig_im_size"],
             }
             masks, scores, logits = self.ort_session.run(None, ort_inputs)
+            masks = masks > self.predictor.model.mask_threshold
+
         else:
             raise ("Not implement now!")
         # masks (n, h, w), scores (n,), logits (n, 256, 256)
-        return masks, scores, logits
+        return masks[0], scores[0], logits[0]
diff --git a/utils/interact_tools.py b/utils/interact_tools.py
@@ -23,6 +23,7 @@ def __init__(self, sam_pt_checkpoint, sam_onnx_checkpoint, model_type, device):
         """
 
         self.sam_controler = BaseSegmenter(sam_pt_checkpoint, sam_onnx_checkpoint, model_type, device)
+        self.onnx = model_type == "vit_t"
 
     def first_frame_click(
         self,
@@ -38,32 +39,38 @@ def first_frame_click(
         """
         # self.sam_controler.set_image(image)
         neg_flag = labels[-1]
-        if neg_flag == 1:
-            # find neg
+
+        if self.onnx:
+            onnx_coord = np.concatenate([points, np.array([[0.0, 0.0]])], axis=0)[None, :, :]
+            onnx_label = np.concatenate([labels, np.array([-1])], axis=0)[None, :].astype(np.float32)
+            onnx_coord = self.sam_controler.predictor.transform.apply_coords(onnx_coord, image.shape[:2]).astype(np.float32)
+            prompts = {
+                "point_coords": onnx_coord,
+                "point_labels": onnx_label,
+                "orig_im_size": np.array(image.shape[:2], dtype=np.float32),
+            }
+
+        else:
             prompts = {
                 "point_coords": points,
                 "point_labels": labels,
-                "orig_im_size": image.shape[:2],
             }
+
+        if neg_flag == 1:
+            # find positive
             masks, scores, logits = self.sam_controler.predict(
                 prompts, "point", multimask
             )
             mask, logit = masks[np.argmax(scores)], logits[np.argmax(scores), :, :]
-            prompts = {
-                "point_coords": points,
-                "point_labels": labels,
-                "mask_input": logit[None, :, :],
-            }
+
+            prompts["mask_input"] = np.expand_dims(logit[None, :, :], 0)
             masks, scores, logits = self.sam_controler.predict(
                 prompts, "both", multimask
             )
             mask, logit = masks[np.argmax(scores)], logits[np.argmax(scores), :, :]
+
         else:
-            # find positive
-            prompts = {
-                "point_coords": points,
-                "point_labels": labels,
-            }
+            # find neg
             masks, scores, logits = self.sam_controler.predict(
                 prompts, "point", multimask
             )