remove some magic number

Turoad · Turoad · Sep 11, 2021 · Jun 8, 2021 · Jun 19, 2021 · Jun 19, 2021
commit 9faf0dc46834ba47e6fbd092304a7141ae1b154e
diff --git a/configs/condlane/resnet101_culane.py b/configs/condlane/resnet101_culane.py
@@ -11,8 +11,6 @@
     in_channels=[64, 128, 256, 512]
 )
 
-featuremap_out_channel = 128
-featuremap_out_stride = 8
 sample_y = range(589, 230, -20)
 
 batch_size = 8
@@ -95,6 +93,7 @@
 nms_thr = 4
 img_scale = (800, 320)
 crop_bbox = [0, 270, 1640, 590]
+mask_size = (1, 80, 200)
 
 train_process = [
     dict(type='Alaug',

diff --git a/lanedet/datasets/culane.py b/lanedet/datasets/culane.py
@@ -52,13 +52,10 @@ def load_annotation(self, line):
         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 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
 
         return infos
 

diff --git a/lanedet/engine/runner.py b/lanedet/engine/runner.py
@@ -110,6 +110,7 @@ def validate(self):
             data = self.to_cuda(data)
             with torch.no_grad():
                 output = self.net(data)
+                output = self.net.module.get_lanes(output)
                 predictions.extend(output)
             if self.cfg.view:
                 self.val_loader.dataset.view(output, data['meta'])

diff --git a/lanedet/models/heads/condlane.py b/lanedet/models/heads/condlane.py
@@ -705,9 +705,8 @@ def __init__(self,
         self.feat_width = location_configs['size'][-1]
         self.mlp = MLP(self.feat_width, 64, 2, 2)
 
-        # TODO(check)
         self.post_process = CondLanePostProcessor(
-                mask_size=(1, 80, 200), hm_thr=0.5, use_offset=True,
+                mask_size=self.cfg.mask_size, hm_thr=0.5, use_offset=True,
                 nms_thr=4)
 
         self.loss_impl = CondLaneLoss(cfg.loss_weights, 1)
@@ -749,10 +748,6 @@ def parse_gt(self, gts, device):
             gts['row']).to(device)).unsqueeze(0).unsqueeze(0)
         row_mask = (torch.from_numpy(
             gts['row_mask']).to(device)).unsqueeze(0).unsqueeze(0)
-        # reg = (gts['reg']).to(device)#.unsqueeze(0)
-        # reg_mask = (gts['reg_mask']).to(device)#.unsqueeze(0)
-        # row = (gts['row']).to(device).unsqueeze(0)#.unsqueeze(0)
-        # row_mask = (gts['row_mask']).to(device).unsqueeze(0)#.unsqueeze(0)
         if 'range' in gts:
             lane_range = torch.from_numpy(gts['range']).to(device) # new add: squeeze 
             #lane_range = (gts['range']).to(device).squeeze(0) # new add: squeeze 
@@ -782,19 +777,12 @@ def parse_pos(self, gt_masks, hm_shape, device, mask_shape=None):
             for m in m_img:
                 gts = self.parse_gt(m, device=device)
                 reg, reg_mask, row, row_mask, lane_range = gts
-                # print('reg: ', reg.shape)
-                # print('reg_mask: ', reg_mask.shape)
-                # print('row ns:', row.shape)
-                # print('row_mask:', row_mask.shape)
-                # print('lane_range:', lane_range.shape)
                 label = m['label']
                 num += len(m['points'])
                 for p in m['points']:
                     pos = idx * n * hm_h * hm_w + label * hm_h * hm_w + p[
                         1] * hm_w + p[0]
-                    # pos = [idx, label, p[1], p[0]]
                     poses.append(pos)
-                # m['label'] = torch.from_numpy(np.array(m['label'])).to(device)
                 for i in range(len(m['points'])):
                     labels.append(label)
                     regs.append(reg)
@@ -878,8 +866,8 @@ def _format(heat, inds):
     def forward_train(self, output, batch):
         img_metas = batch['img_metas']
         gt_batch_masks = [m['gt_masks'] for m in img_metas]
-        hm_shape = [20, 50]
-        mask_shape = [80, 200]
+        hm_shape = img_metas[0]['hm_shape']
+        mask_shape = img_metas[0]['mask_shape']
         inputs = output
         pos, labels, num_ins, gts = self.parse_pos(
             gt_batch_masks, hm_shape, inputs[0].device, mask_shape=mask_shape)
@@ -1025,17 +1013,13 @@ def forward(
 
     def get_lanes(self, output):
         out_seeds, out_hm = output['seeds'], output['hm']
-        #TODO
         ret = []
         for seeds, hm in zip(out_seeds, out_hm):
-            downscale=4
-            lanes, seed = self.post_process(seeds, downscale)
-            crop_bbox = [0, 270, 1640, 590]
-            img_shape = (320, 800, 3) 
+            lanes, seed = self.post_process(seeds, self.cfg.mask_down_scale)
             result = adjust_result(
                     lanes=lanes,
-                    crop_bbox=crop_bbox,
-                    img_shape=img_shape,
+                    crop_bbox=self.cfg.crop_bbox,
+                    img_shape=self.cfg.img_scale,
                     tgt_shape=(590, 1640),
                     )
             lanes = []
@@ -1044,7 +1028,6 @@ def get_lanes(self, output):
                 for x, y in lane:
                     coord.append([x, y])
                 coord = np.array(coord)
-                #coord = np.flip(coord, axis=0)
                 coord[:, 0] /= self.cfg.ori_img_w
                 coord[:, 1] /= self.cfg.ori_img_h
                 lanes.append(Lane(coord))

diff --git a/lanedet/models/necks/fpn.py b/lanedet/models/necks/fpn.py
@@ -110,7 +110,11 @@ def __init__(self,
 
     def forward(self, inputs):
         """Forward function."""
-        assert len(inputs) == len(self.in_channels)
+        assert len(inputs) >= len(self.in_channels)
+
+        if len(inputs) > len(self.in_channels):
+            for _ in range(len(inputs) - len(self.in_channels)):
+                del inputs[0]
 
         # build laterals
         laterals = [

diff --git a/lanedet/models/nets/detector.py b/lanedet/models/nets/detector.py
@@ -33,6 +33,5 @@ def forward(self, batch):
             output.update(self.heads.loss(out, batch))
         else:
             output = self.heads(fea)
-            output = self.heads.get_lanes(output)
 
         return output
diff --git a/tools/detect.py b/tools/detect.py
@@ -25,7 +25,7 @@ def __init__(self, cfg):
     def preprocess(self, img_path):
         ori_img = cv2.imread(img_path)
         img = ori_img[self.cfg.cut_height:, :, :].astype(np.float32)
-        data = {'img': img}
+        data = {'img': img, 'lanes': []}
         data = self.processes(data)
         data['img'] = data['img'].unsqueeze(0)
         data.update({'img_path':img_path, 'ori_img':ori_img})