[Fix] Fix CI error due to np.int and legacy builder.py

.circleci/test.yml

@@ -66,9 +66,9 @@ jobs:
           name: Install mmyolo dependencies
           command: |
             pip install -U openmim
-            mim install 'mmengine >= 0.3.1'
+            mim install git+https://github.com/open-mmlab/mmengine.git@main
             mim install 'mmcv >= 2.0.0rc1'
-            pip install git+https://github.com/open-mmlab/[email protected]
+            mim install git+https://github.com/open-mmlab/[email protected]
             pip install -r requirements/albu.txt
             pip install -r requirements/tests.txt
       - run:

docs/en/algorithm_descriptions/yolov5_description.md

@@ -618,7 +618,7 @@ At the same time, the batch shape of the current batch is calculated to prevent
 
         n = len(image_shapes)  # number of images
         batch_index = np.floor(np.arange(n) / self.batch_size).astype(
-            np.int)  # batch index
+            np.int64)  # batch index
         number_of_batches = batch_index[-1] + 1  # number of batches
 
         aspect_ratio = image_shapes[:, 1] / image_shapes[:, 0]  # aspect ratio
@@ -640,7 +640,7 @@ At the same time, the batch shape of the current batch is calculated to prevent
 
         batch_shapes = np.ceil(
             np.array(shapes) * self.img_size / self.size_divisor +
-            self.pad).astype(np.int) * self.size_divisor
+            self.pad).astype(np.int64) * self.size_divisor
 
         for i, data_info in enumerate(data_list):
             data_info['batch_shape'] = batch_shapes[batch_index[i]]

docs/zh_cn/algorithm_descriptions/yolov5_description.md

@@ -619,7 +619,7 @@ YOLOv5 是非解耦输出头，而其他大部分算法都是解耦输出头，
 
         n = len(image_shapes)  # number of images
         batch_index = np.floor(np.arange(n) / self.batch_size).astype(
-            np.int)  # batch index
+            np.int64)  # batch index
         number_of_batches = batch_index[-1] + 1  # number of batches
 
         aspect_ratio = image_shapes[:, 1] / image_shapes[:, 0]  # aspect ratio
@@ -641,7 +641,7 @@ YOLOv5 是非解耦输出头，而其他大部分算法都是解耦输出头，
 
         batch_shapes = np.ceil(
             np.array(shapes) * self.img_size / self.size_divisor +
-            self.pad).astype(np.int) * self.size_divisor
+            self.pad).astype(np.int64) * self.size_divisor
 
         for i, data_info in enumerate(data_list):
             data_info['batch_shape'] = batch_shapes[batch_index[i]]

mmyolo/datasets/utils.py

@@ -64,7 +64,7 @@ def __call__(self, data_list: List[dict]) -> List[dict]:
 
         n = len(image_shapes)  # number of images
         batch_index = np.floor(np.arange(n) / self.batch_size).astype(
-            np.int)  # batch index
+            np.int64)  # batch index
         number_of_batches = batch_index[-1] + 1  # number of batches
 
         aspect_ratio = image_shapes[:, 1] / image_shapes[:, 0]  # aspect ratio
@@ -86,7 +86,7 @@ def __call__(self, data_list: List[dict]) -> List[dict]:
 
         batch_shapes = np.ceil(
             np.array(shapes) * self.img_size / self.size_divisor +
-            self.extra_pad_ratio).astype(np.int) * self.size_divisor
+            self.extra_pad_ratio).astype(np.int64) * self.size_divisor
 
         for i, data_info in enumerate(data_list):
             data_info['batch_shape'] = batch_shapes[batch_index[i]]

mmyolo/utils/boxam_utils.py

@@ -12,13 +12,14 @@
 import torchvision
 from mmcv.transforms import Compose
 from mmdet.evaluation import get_classes
-from mmdet.models import build_detector
 from mmdet.utils import ConfigType
 from mmengine.config import Config
 from mmengine.runner import load_checkpoint
 from mmengine.structures import InstanceData
 from torch import Tensor
 
+from mmyolo.registry import MODELS
+
 try:
     from pytorch_grad_cam import (AblationCAM, AblationLayer,
                                   ActivationsAndGradients)
@@ -71,7 +72,7 @@ def init_detector(
     # grad based method requires train_cfg
     # config.model.train_cfg = None
 
-    model = build_detector(config.model)
+    model = MODELS.build(config.model)
     if checkpoint is not None:
         checkpoint = load_checkpoint(model, checkpoint, map_location='cpu')
         # Weights converted from elsewhere may not have meta fields.

tests/test_models/test_detectors/test_yolo_detector.py

@@ -29,8 +29,8 @@ def test_init(self, cfg_file):
         model = get_detector_cfg(cfg_file)
         model.backbone.init_cfg = None
 
-        from mmdet.models import build_detector
-        detector = build_detector(model)
+        from mmyolo.registry import MODELS
+        detector = MODELS.build(model)
         self.assertTrue(detector.backbone)
         self.assertTrue(detector.neck)
         self.assertTrue(detector.bbox_head)
@@ -56,11 +56,11 @@ def test_forward_loss_mode(self, cfg_file, devices):
                 std=[255., 255., 255.],
                 bgr_to_rgb=True)
 
-        from mmdet.models import build_detector
+        from mmyolo.registry import MODELS
         assert all([device in ['cpu', 'cuda'] for device in devices])
 
         for device in devices:
-            detector = build_detector(model)
+            detector = MODELS.build(model)
             detector.init_weights()
 
             if device == 'cuda':
@@ -85,11 +85,11 @@ def test_forward_predict_mode(self, cfg_file, devices):
         model = get_detector_cfg(cfg_file)
         model.backbone.init_cfg = None
 
-        from mmdet.models import build_detector
+        from mmyolo.registry import MODELS
         assert all([device in ['cpu', 'cuda'] for device in devices])
 
         for device in devices:
-            detector = build_detector(model)
+            detector = MODELS.build(model)
 
             if device == 'cuda':
                 if not torch.cuda.is_available():
@@ -117,11 +117,11 @@ def test_forward_tensor_mode(self, cfg_file, devices):
         model = get_detector_cfg(cfg_file)
         model.backbone.init_cfg = None
 
-        from mmdet.models import build_detector
+        from mmyolo.registry import MODELS
         assert all([device in ['cpu', 'cuda'] for device in devices])
 
         for device in devices:
-            detector = build_detector(model)
+            detector = MODELS.build(model)
 
             if device == 'cuda':
                 if not torch.cuda.is_available():

tools/analysis_tools/benchmark.py

@@ -5,14 +5,14 @@
 import time
 
 import torch
-from mmdet.models import build_detector
 from mmengine import Config, DictAction
 from mmengine.dist import get_world_size, init_dist
 from mmengine.logging import MMLogger, print_log
 from mmengine.runner import Runner, load_checkpoint
 from mmengine.utils import mkdir_or_exist
 from mmengine.utils.dl_utils import set_multi_processing
 
+from mmyolo.registry import MODELS
 from mmyolo.utils import register_all_modules
 
 register_all_modules()
@@ -82,7 +82,7 @@ def measure_inference_speed(cfg, checkpoint, max_iter, log_interval,
     data_loader = Runner.build_dataloader(dataloader_cfg)
 
     # build the model and load checkpoint
-    model = build_detector(cfg.model)
+    model = MODELS.build(cfg.model)
     load_checkpoint(model, checkpoint, map_location='cpu')
     model = model.cuda()
     model.eval()