Release model: r101_nuimg_1408x512

README.md

@@ -4,14 +4,15 @@ This is the official PyTorch implementation for paper [SparseBEV: High-Performan
 
 ## Model Zoo
 
-| Setting  | Pretrain | Training Cost | NDS | FPS | Config | Weights |
-|----------|----------|---------------|-----|-----|--------|---------|
-| r50_nuimg_704x256 | [nuImages](https://download.openmmlab.com/mmdetection3d/v0.1.0_models/nuimages_semseg/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim_20201009_124951-40963960.pth) | 21h (8x2080Ti) | 55.6 | 15.8 | [config](configs/r50_nuimg_704x256.py) | [weights](https://drive.google.com/file/d/1ft34-pxLpHGo2Aw-jowEtCxyXcqszHNn/view) |
-| r50_nuimg_704x256_400q_36ep | [nuImages](https://download.openmmlab.com/mmdetection3d/v0.1.0_models/nuimages_semseg/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim_20201009_124951-40963960.pth) | 28h (8x2080Ti) | 55.8 | 23.5 | [config](configs/r50_nuimg_704x256_400q_36ep.py) | [weights](https://drive.google.com/file/d/1C_Vn3iiSnSW1Dw1r0DkjJMwvHC5Y3zTN/view) |
+| Setting  | Pretrain | Training Cost | NDS | FPS | Weights |
+|----------|----------|---------------|-----|-----|---------|
+| [r50_nuimg_704x256](configs/r50_nuimg_704x256.py) | [nuImages](https://download.openmmlab.com/mmdetection3d/v0.1.0_models/nuimages_semseg/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim_20201009_124951-40963960.pth) | 21h (8x2080Ti) | 55.6 | 15.8 | [weights](https://drive.google.com/file/d/1ft34-pxLpHGo2Aw-jowEtCxyXcqszHNn/view) |
+| [r50_nuimg_704x256_400q_36ep](configs/r50_nuimg_704x256_400q_36ep.py) | [nuImages](https://download.openmmlab.com/mmdetection3d/v0.1.0_models/nuimages_semseg/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim/cascade_mask_rcnn_r50_fpn_coco-20e_20e_nuim_20201009_124951-40963960.pth) | 28h (8x2080Ti) | 55.8 | 23.5 | [weights](https://drive.google.com/file/d/1C_Vn3iiSnSW1Dw1r0DkjJMwvHC5Y3zTN/view) |
+| [r101_nuimg_1408x512](configs/r101_nuimg_1408x512.py) | [nuImages](https://download.openmmlab.com/mmdetection3d/v0.1.0_models/nuimages_semseg/cascade_mask_rcnn_r101_fpn_1x_nuim/cascade_mask_rcnn_r101_fpn_1x_nuim_20201024_134804-45215b1e.pth) | 2d8h (8xV100) | 59.2 | 6.5 | [weights](https://drive.google.com/file/d/1dKu5cR1fuo-O0ynyBh-RCPtHrgut29mN/view) |
 
 * We use `r50_nuimg_704x256` for ablation studies and `r50_nuimg_704x256_400q_36ep` for comparison with others.
 * We recommend using `r50_nuimg_704x256` to validate new ideas since it trains faster and the result is more stable.
-* FPS is measured with AMD 5800X CPU and RTX 3090 GPU.
+* FPS is measured with AMD 5800X CPU and RTX 3090 GPU (without `fp16`).
 * The noise is around 0.3 NDS.
 
 ## Environment
@@ -107,14 +108,14 @@ Single-GPU evaluation:
 
 ```
 export CUDA_VISIBLE_DEVICES=0
-python val.py --config configs/r50_nuimg_704x256_400q_36ep.py --weights checkpoints/r50_nuimg_704x256_400q_36ep.pth
+python val.py --config configs/r50_nuimg_704x256.py --weights checkpoints/r50_nuimg_704x256.pth
 ```
 
 Multi-GPU evaluation:
 
 ```
 export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
-torchrun --nproc_per_node 8 val.py --config configs/r50_nuimg_704x256_400q_36ep.py --weights checkpoints/r50_nuimg_704x256_400q_36ep.pth
+torchrun --nproc_per_node 8 val.py --config configs/r50_nuimg_704x256.py --weights checkpoints/r50_nuimg_704x256.pth
 ```
 
 ## Timing
@@ -123,7 +124,7 @@ FPS is measured with a single GPU:
 
 ```
 export CUDA_VISIBLE_DEVICES=0
-python timing.py --config configs/r50_nuimg_704x256_400q_36ep.py --weights checkpoints/r50_nuimg_704x256_400q_36ep.pth
+python timing.py --config configs/r50_nuimg_704x256.py --weights checkpoints/r50_nuimg_704x256.pth
 ```
 
 ## Acknowledgements

models/sparsebev_head.py

@@ -8,6 +8,7 @@ from mmdet.models.dense_heads import DETRHead
 from mmdet3d.core.bbox.coders import build_bbox_coder
 from mmdet3d.core.bbox.structures.lidar_box3d import LiDARInstance3DBoxes
 from .bbox.utils import normalize_bbox, encode_bbox
+from .utils import VERSION
 
 
 @HEADS.register_module()
@@ -462,6 +463,13 @@ class SparseBEVHead(DETRHead):
             preds = preds_dicts[i]
             bboxes = preds['bboxes']
             bboxes[:, 2] = bboxes[:, 2] - bboxes[:, 5] * 0.5
+
+            if VERSION.name == 'v0.17.1':
+                import copy
+                w, l = copy.deepcopy(bboxes[:, 3]), copy.deepcopy(bboxes[:, 4])
+                bboxes[:, 3], bboxes[:, 4] = l, w
+                bboxes[:, 6] = -bboxes[:, 6] - math.pi / 2
+
             bboxes = LiDARInstance3DBoxes(bboxes, 9)
             scores = preds['scores']
             labels = preds['labels']

models/utils.py

@@ -62,11 +62,18 @@ def rotation_3d_in_axis(points, angles):
     ones = torch.ones_like(rot_cos)
     zeros = torch.zeros_like(rot_cos)
 
-    rot_mat_T = torch.stack([
-        rot_cos, rot_sin, zeros,
-        -rot_sin, rot_cos, zeros,
-        zeros, zeros, ones,
-    ]).transpose(0, 1).reshape(-1, 3, 3)
+    if VERSION.name == 'v0.17.1':
+        rot_mat_T = torch.stack([
+            rot_cos, -rot_sin, zeros,
+            rot_sin, rot_cos, zeros,
+            zeros, zeros, ones,
+        ]).transpose(0, 1).reshape(-1, 3, 3)
+    else:
+        rot_mat_T = torch.stack([
+            rot_cos, rot_sin, zeros,
+            -rot_sin, rot_cos, zeros,
+            zeros, zeros, ones,
+        ]).transpose(0, 1).reshape(-1, 3, 3)
 
     points = torch.bmm(points, rot_mat_T)
 
@@ -307,3 +314,11 @@ class DumpConfig:
 
 
 DUMP = DumpConfig()
+
+
+# for backward compatibility
+class Version:
+    def __init__(self):
+        self.name = 'v1.0.0'
+
+VERSION = Version()

val.py

@@ -13,6 +13,7 @@ from mmcv.runner import load_checkpoint
 from mmdet.apis import set_random_seed, multi_gpu_test, single_gpu_test
 from mmdet3d.datasets import build_dataset, build_dataloader
 from mmdet3d.models import build_model
+from models.utils import VERSION
 
 
 def evaluate(dataset, results, epoch):
@@ -117,20 +118,23 @@ def main():
     else:
         model = MMDataParallel(model, [0])
 
-    if os.path.isfile(args.weights):
-        logging.info('Loading checkpoint from %s' % args.weights)
-        load_checkpoint(
-            model, args.weights, map_location='cuda', strict=True,
-            logger=logging.Logger(__name__, logging.ERROR)
-        )
+    logging.info('Loading checkpoint from %s' % args.weights)
+    checkpoint = load_checkpoint(
+        model, args.weights, map_location='cuda', strict=True,
+        logger=logging.Logger(__name__, logging.ERROR)
+    )
 
-        if world_size > 1:
-            results = multi_gpu_test(model, val_loader, gpu_collect=True)
-        else:
-            results = single_gpu_test(model, val_loader)
+    if 'version' in checkpoint:
+        VERSION.name = checkpoint['version']
+        logging.info(VERSION.name)
 
-        if local_rank == 0:
-            evaluate(val_dataset, results, -1)
+    if world_size > 1:
+        results = multi_gpu_test(model, val_loader, gpu_collect=True)
+    else:
+        results = single_gpu_test(model, val_loader)
+
+    if local_rank == 0:
+        evaluate(val_dataset, results, -1)
 
 
 if __name__ == '__main__':