ef35a38ad29c@2023-12-07_15-44-23:

shenyunhang · Dec 7, 2023 · 0bc4795 · 0bc4795
1 parent a19a5b2
commit 0bc4795
Show file tree

Hide file tree

Showing 5 changed files with 33 additions and 28 deletions.
diff --git a/ape/modeling/ape_deta/deformable_detr_segm.py b/ape/modeling/ape_deta/deformable_detr_segm.py
@@ -248,9 +248,9 @@ def forward(self, batched_inputs, do_postprocess=True):
                 text_list = text_list[:1203+365+601]
                 cache = True
 
-                from detectron2.data.catalog import MetadataCatalog
-                metadata = MetadataCatalog.get("coco_2017_train_panoptic_separated")
-                text_list = get_text_list(metadata, "thing+stuff")
+                # from detectron2.data.catalog import MetadataCatalog
+                # metadata = MetadataCatalog.get("coco_2017_train_panoptic_separated")
+                # text_list = get_text_list(metadata, "thing+stuff")
 
             outputs_l = self.model_language.forward_text(text_list, cache=cache)
             if "last_hidden_state_eot" in outputs_l:

diff --git a/ape/modeling/ape_deta/deformable_detr_segm_vl.py b/ape/modeling/ape_deta/deformable_detr_segm_vl.py
@@ -248,9 +248,9 @@ def forward(self, batched_inputs, do_postprocess=True):
                 text_list = text_list[:1203+365+601]
                 cache = True
 
-                from detectron2.data.catalog import MetadataCatalog
-                metadata = MetadataCatalog.get("coco_2017_train_panoptic_separated")
-                text_list = get_text_list(metadata, "thing+stuff")
+                # from detectron2.data.catalog import MetadataCatalog
+                # metadata = MetadataCatalog.get("coco_2017_train_panoptic_separated")
+                # text_list = get_text_list(metadata, "thing+stuff")
 
             outputs_l = self.model_language.forward_text(text_list, cache=cache)
             if "last_hidden_state_eot" in outputs_l:

diff --git a/demo/app.py b/demo/app.py
@@ -24,6 +24,9 @@
 
 this_dir = path.dirname(path.abspath(__file__))
 
+os.system("git clone https://github.com/shenyunhang/APE.git")
+os.system("python3.10 -m pip install -e APE/")
+
 example_list = [
     [
         this_dir + "/examples/Totoro01.png",
@@ -292,11 +295,11 @@ def load_APE_A():
         "train.init_checkpoint='{}'".format(init_checkpoint),
         "model.model_language.cache_dir=''",
         "model.model_vision.select_box_nums_for_evaluation=500",
+        "model.model_vision.backbone.net.xattn=False",
     ]
     if running_device == "cpu":
         args.opts += [
             "model.model_language.dtype='float32'",
-            "model.model_vision.backbone.net.xattn=False",
         ]
     logger.info("Arguments: " + str(args))
     cfg = setup_cfg(args)
@@ -338,11 +341,11 @@ def load_APE_B():
         "model.model_language.cache_dir=''",
         "model.model_vision.select_box_nums_for_evaluation=500",
         "model.model_vision.text_feature_bank_reset=True",
+        "model.model_vision.backbone.net.xattn=False",
     ]
     if running_device == "cpu":
         args.opts += [
             "model.model_language.dtype='float32'",
-            "model.model_vision.backbone.net.xattn=False",
         ]
     logger.info("Arguments: " + str(args))
     cfg = setup_cfg(args)
@@ -384,11 +387,11 @@ def load_APE_C():
         "model.model_language.cache_dir=''",
         "model.model_vision.select_box_nums_for_evaluation=500",
         "model.model_vision.text_feature_bank_reset=True",
+        "model.model_vision.backbone.net.xattn=False",
     ]
     if running_device == "cpu":
         args.opts += [
             "model.model_language.dtype='float32'",
-            "model.model_vision.backbone.net.xattn=False",
         ]
     logger.info("Arguments: " + str(args))
     cfg = setup_cfg(args)
@@ -430,11 +433,11 @@ def load_APE_D():
         "model.model_language.cache_dir=''",
         "model.model_vision.select_box_nums_for_evaluation=500",
         "model.model_vision.text_feature_bank_reset=True",
+        "model.model_vision.backbone.net.xattn=False",
     ]
     if running_device == "cpu":
         args.opts += [
             "model.model_language.dtype='float32'",
-            "model.model_vision.backbone.net.xattn=False",
         ]
     logger.info("Arguments: " + str(args))
     cfg = setup_cfg(args)
@@ -746,9 +749,10 @@ def add_tail_info():
     all_demo = {}
     all_cfg = {}
 
-    load_APE_A()
-    load_APE_B()
-    load_APE_C()
+    # load_APE_A()
+    # load_APE_B()
+    # load_APE_C()
+    save_memory = False
     load_APE_D()
 
     title = "APE: Aligning and Prompting Everything All at Once for Universal Visual Perception"

diff --git a/demo/pre-requirements.txt b/demo/pre-requirements.txt
@@ -3,3 +3,4 @@ torch==1.12.1
 torchvision
 transformers==4.32.1
 scipy
+cython
diff --git a/demo/predictor_lazy.py b/demo/predictor_lazy.py
@@ -150,21 +150,21 @@ def __init__(self, cfg, instance_mode=ColorMode.IMAGE, parallel=False, args=None
             + MetadataCatalog.get(d).get("stuff_classes", default=["thing"])[1:]
         ]
 
-        self.metadata = MetadataCatalog.get(
-            "__unused_ape_" + "_".join([d for d in cfg.dataloader.train.dataset.names])
-        )
-        self.metadata.thing_classes = [
-            c
-            for d in ["coco_2017_train_panoptic_separated"]
-            for c in MetadataCatalog.get(d).get("thing_classes", default=[])
-            + MetadataCatalog.get(d).get("stuff_classes", default=["thing"])[1:]
-        ]
-        self.metadata.stuff_classes = [
-            c
-            for d in ["coco_2017_train_panoptic_separated"]
-            for c in MetadataCatalog.get(d).get("thing_classes", default=[])
-            + MetadataCatalog.get(d).get("stuff_classes", default=["thing"])[1:]
-        ]
+        # self.metadata = MetadataCatalog.get(
+        #     "__unused_ape_" + "_".join([d for d in cfg.dataloader.train.dataset.names])
+        # )
+        # self.metadata.thing_classes = [
+        #     c
+        #     for d in ["coco_2017_train_panoptic_separated"]
+        #     for c in MetadataCatalog.get(d).get("thing_classes", default=[])
+        #     + MetadataCatalog.get(d).get("stuff_classes", default=["thing"])[1:]
+        # ]
+        # self.metadata.stuff_classes = [
+        #     c
+        #     for d in ["coco_2017_train_panoptic_separated"]
+        #     for c in MetadataCatalog.get(d).get("thing_classes", default=[])
+        #     + MetadataCatalog.get(d).get("stuff_classes", default=["thing"])[1:]
+        # ]
 
         self.cpu_device = torch.device("cpu")
         self.instance_mode = instance_mode
-Original file line number
+Diff line change
@@ Expand Up / @@ -3,3 +3,4 @@ torch==1.12.1 @@
     torchvision
     transformers==4.32.1
     scipy
+    cython