update condlane

Turoad · Turoad · Sep 11, 2021 · Jun 8, 2021 · Jun 19, 2021 · Jun 19, 2021
commit 8d6e0227a0b00322349ab397441dcbfa94a2a784
diff --git a/configs/condlane/resnet101_culane.py b/configs/condlane/resnet101_culane.py
@@ -61,7 +61,7 @@
     location_configs=dict(size=(batch_size, 1, 80, 200), device='cuda:0')
 )
 
-optimizer = dict(type='Adam', lr=6e-4, betas=(0.9, 0.999), eps=1e-8)
+optimizer = dict(type='AdamW', lr=3e-4, betas=(0.9, 0.999), eps=1e-8)
 
 epochs = 16
 total_iter = (88880 // batch_size) * epochs

diff --git a/lanedet/datasets/culane.py b/lanedet/datasets/culane.py
@@ -49,18 +49,17 @@ def load_annotation(self, line):
         anno_path = img_path[:-3] + 'lines.txt'  # remove sufix jpg and add lines.txt
         with open(anno_path, 'r') as anno_file:
             data = [list(map(float, line.split())) for line in anno_file.readlines()]
-        # lanes = [[(lane[i], lane[i + 1]) for i in range(0, len(lane), 2) if lane[i] >= 0 and lane[i + 1] >= 0]
-        #          for lane in data]
-        # lanes = [list(set(lane)) for lane in lanes]  # remove duplicated points
-        # lanes = [lane for lane in lanes if len(lane) >= 2]  # remove lanes with less than 2 points
+        lanes = [[(lane[i], lane[i + 1]) for i in range(0, len(lane), 2) if lane[i] >= 0 and lane[i + 1] >= 0]
+                 for lane in data]
+        lanes = [list(set(lane)) for lane in lanes]  # remove duplicated points
+        lanes = [lane for lane in lanes if len(lane) > 3]  # remove lanes with less than 2 points
 
-        # lanes = [sorted(lane, key=lambda x: x[1]) for lane in lanes]  # sort by y
-        gt_points = [[float(lane[i]) for i in range(len(lane))] for lane in data]
-        gt_points = [gt_point for gt_point in gt_points if len(gt_point) > 3]
-        #infos['lanes'] = lanes
-        infos['gt_points'] = gt_points
-        infos['id_classes'] = [1 for i in range(len(gt_points))]
-        infos['id_instances'] = [i + 1 for i in range(len(gt_points))]
+        lanes = [sorted(lane, key=lambda x: x[1]) for lane in lanes]  # sort by y
+        # gt_points = [[float(lane[i]) for i in range(len(lane))] for lane in data]
+        # gt_points = [gt_point for gt_point in gt_points if len(gt_point) > 3]
+
+        infos['lanes'] = lanes
+        # infos['gt_points'] = gt_points
 
         return infos
 
@@ -69,11 +68,6 @@ def get_prediction_string(self, pred):
         ys = np.arange(270, 590, 8) / self.cfg.ori_img_h
         out = []
         for lane in pred:
-            # points = lane.points
-            # lane_str = ' '.join(['{:.5f} {:.5f}'.format(x, y) for x, y in points])
-            # if lane_str != '':
-            #     out.append(lane_str)
-            # continue
             xs = lane(ys)
             valid_mask = (xs >= 0) & (xs < 1)
             xs = xs * self.cfg.ori_img_w

diff --git a/lanedet/datasets/process/alaug.py b/lanedet/datasets/process/alaug.py
@@ -140,12 +140,21 @@ def __call__(self, data):
             else:
                 keypoints_val = keypoints_val + points_val
 
+        if 'lanes' in data:
+            points_val = []
+            for lane in data['lanes']:
+                points_val.extend(lane)
+
+            points_index = [len(lane) for lane in data['lanes']]
+            keypoints_val = points_val
+
         aug = self.__augmentor(
             image=img,
             keypoints=keypoints_val,
             bboxes=bboxes,
             mask=masks,
             bbox_labels=bbox_labels)
+
         data['img'] = aug['image']
         data['img_shape'] = data['img'].shape
         if 'gt_bboxes' in data:
@@ -156,23 +165,8 @@ def __call__(self, data):
                 return None
         if 'gt_masks' in data:
             data['gt_masks'] = [np.array(aug['mask'])]
-        if 'gt_keypoints' in data:
-            kp_list = [[0 for j in range(i * 2)] for i in keypoints_index]
-            for i in range(len(keypoints_index)):
-                for j in range(keypoints_index[i]):
-                    kp_list[i][2 * j] = aug['keypoints'][
-                        self.cal_sum_list(keypoints_index, i) + j][0]
-                    kp_list[i][2 * j + 1] = aug['keypoints'][
-                        self.cal_sum_list(keypoints_index, i) + j][1]
-            data['gt_keypoints'] = []
-            valid = []
-            for i in range(kp_group_num):
-                index = int(aug['bboxes'][i][-1])
-                valid.append(index)
-                data['gt_keypoints'].append(kp_list[index])
-            data['gt_keypoints_ignore'] = data['gt_keypoints_ignore'][valid]
 
-        if 'gt_points' in data:
+        if 'gt_points' in data or 'lanes' in data:
             start_idx = num_keypoints if 'gt_keypoints' in data else 0
             points = aug['keypoints'][start_idx:]
             kp_list = [[0 for j in range(i * 2)] for i in points_index]

diff --git a/lanedet/datasets/process/collect_lane.py b/lanedet/datasets/process/collect_lane.py
@@ -228,9 +228,12 @@ def min_dis_one_point(points, idx):
 
         # gt heatmap and ins of bank
         gt_points = results['gt_points']
+        # gt_points = results['lanes']
         valid_gt = []
         for pts in gt_points:
             id_class = 1
+            # pts = np.array(pts) / self.down_scale
+            #pts = [(pt[0]/self.down_scale, pt[1]/self.down_scale) for pt in pts]
             pts = convert_list(pts, self.down_scale)
             pts = sorted(pts, key=cmp_to_key(lambda a, b: b[-1] - a[-1]))
             pts = clamp_line(
@@ -240,21 +243,20 @@ def min_dis_one_point(points, idx):
 
         # draw gt_hm_lane
         gt_hm_lane_ends = []
-        radius = []
         for l in valid_gt:
             label = l[1]
             point = (l[0][0][0] * ratio_hm_mask, l[0][0][1] * ratio_hm_mask)
             gt_hm_lane_ends.append([point, l[0]])
-        for i, p in enumerate(gt_hm_lane_ends):
-            r = self.radius
-            radius.append(r)
-
-        if len(gt_hm_lane_ends) >= 2:
-            endpoints = [p[0] for p in gt_hm_lane_ends]
-            for j in range(len(endpoints)):
-                dis = min_dis_one_point(endpoints, j)
-                if dis < 1.5 * radius[j]:
-                    radius[j] = int(max(dis / 1.5, 1) + 0.49999)
+        radius = [self.radius for _ in range(len(gt_hm_lane_ends))]
+
+        # if len(gt_hm_lane_ends) >= 2:
+        #     endpoints = [p[0] for p in gt_hm_lane_ends]
+        #     for j in range(len(endpoints)):
+        #         dis = min_dis_one_point(endpoints, j)
+        #         if dis < 1.5 * radius[j]:
+        #             raise False
+        #             radius[j] = int(max(dis / 1.5, 1) + 0.49999)
+
         for (end_point, line), r in zip(gt_hm_lane_ends, radius):
             pos = np.zeros((mask_h), np.float32)
             pos_mask = np.zeros((mask_h), np.float32)

diff --git a/lanedet/models/heads/condlane.py b/lanedet/models/heads/condlane.py
@@ -960,9 +960,7 @@ def parse_pos(seeds, batchsize, num_classes, h, w, device):
         h_mask, w_mask = f_mask.size()[2:]
         hms, params = z['hm'], z['params']
         hms = torch.clamp(hms.sigmoid(), min=1e-4, max=1 - 1e-4)
-        print(params.shape)
         params = params.view(m_batchsize, self.num_classes, -1, h_hm, w_hm)
-        print(params.shape)
         # with Timer("Elapsed time in two branch: %f"):  # 0.6ms
         mask_branchs = self.mask_branch(f_mask)
         reg_branchs = mask_branchs