YOLO v3 -- 对自己学习有帮助的资源集锦

一、前言：

学无止境，在学习YOLO v3的理论及使用方法时，有以下文章及github开源代码为我提供了很大帮助，特此附上链接，表示感谢！希望对您也有参考借鉴的意义！

二、学习笔记 / 翻译分析

1. 论文链接：https://pjreddie.com/media/files/papers/YOLOv3.pdf

2. 论文翻译：

3. 论文详解：

4. 整理笔记：

（1） tensorflow yolov3训练自己的数据集，详细教程
（2）【论文解读】Yolo三部曲解读——Yolov3
（3） yolo系列之yolo v3【深度解析】
（4） YOLOv2 论文笔记

三、源码

1.（推荐）https://github.com/YunYang1994/tensorflow-yolov3

四、运行出现问题及解决办法

1. 训练自己的类别（单类别不会出错，多类别会出错）时，加载coco的预训练模型，会报错加载错误。部分报错内容如下：

Error--- Restoring from checkpoint failed. This is most likely due to a mismatch between the current graph and the graph from the checkpoint. Please ensure that you have not altered the graph expected based on the checkpoint. Original error:

2 root error(s) found.
  (0) Invalid argument: Assign requires shapes of both tensors to match. lhs shape= [1,1,512,27] rhs shape= [1,1,512,18]
	 [[node loader_and_saver/save/Assign_93 (defined at D:/personal/new_packages/try/train.py:103) ]]
	 [[loader_and_saver/save/RestoreV2/_74]]
  (1) Invalid argument: Assign requires shapes of both tensors to match. lhs shape= [1,1,512,27] rhs shape= [1,1,512,18]
	 [[node loader_and_saver/save/Assign_93 (defined at D:/personal/new_packages/try/train.py:103) ]]
0 successful operations.
0 derived errors ignored.

解决办法（对单类别/多类别均适用）：

将train.py的108行 self.loader = tf.train.Saver(self.net_var)
改为
variables_to_restore = [v for v in self.net_var if v.name.split('/')[0] not in ['conv_sbbox','conv_mbbox','conv_lbbox']]
self.loader = tf.train.Saver(variables_to_restore)

以上解决办法来自：https://github.com/YunYang1994/tensorflow-yolov3/issues/297，感谢mengqp1994。
解决思路源于train.py的77行-86行：

        with tf.name_scope("define_first_stage_train"):
            self.first_stage_trainable_var_list = []
            for var in tf.trainable_variables():
                var_name = var.op.name
                var_name_mess = str(var_name).split('/')
                if var_name_mess[0] in ['conv_sbbox', 'conv_mbbox', 'conv_lbbox']:
                    self.first_stage_trainable_var_list.append(var)

            first_stage_optimizer = tf.train.AdamOptimizer(self.learn_rate).minimize(self.loss,
                                                                                     var_list=self.first_stage_trainable_var_list)

五、一点个人笔记

1. 在验证部分有一个变量min_overlap，它是用来判断tp和fp的，即所预测的正样本对不对，ovmax的计算也是IOU的思想。(非极大值抑制nms用的是iou_threshold)
   具体代码：

# set minimum overlap
for idx, prediction in enumerate(predictions_data):
         file_id = prediction["file_id"]
         gt_file = tmp_files_path + "/" + file_id + "_ground_truth.json"
         ground_truth_data = json.load(open(gt_file))
         ovmax = -1
         gt_match = -1
         # load prediction bounding-box
         bb = [float(x) for x in prediction["bbox"].split()]
         for obj in ground_truth_data:
             # look for a class_name match
             if obj["class_name"] == class_name:
                 bbgt = [float(x) for x in obj["bbox"].split()]
                 bi = [max(bb[0], bbgt[0]), max(bb[1], bbgt[1]), min(bb[2], bbgt[2]), min(bb[3], bbgt[3])]
                 iw = bi[2] - bi[0] + 1
                 ih = bi[3] - bi[1] + 1
                 if iw > 0 and ih > 0:
                     # compute overlap (IoU) = area of intersection / area of union
                     ua = (bb[2] - bb[0] + 1) * (bb[3] - bb[1] + 1) + (bbgt[2] - bbgt[0]
                                                                       + 1) * (bbgt[3] - bbgt[1] + 1) - iw * ih
                     ov = iw * ih / ua
                     if ov > ovmax:
                         ovmax = ov
                         gt_match = obj

         # set minimum overlap
         min_overlap = MINOVERLAP
         if specific_iou_flagged:
             if class_name in specific_iou_classes:
                 index = specific_iou_classes.index(class_name)
                 min_overlap = float(iou_list[index])
         if ovmax >= min_overlap:
             if "difficult" not in gt_match:
                 if not bool(gt_match["used"]):
                     # true positive
                     tp[idx] = 1
                     gt_match["used"] = True
                     count_true_positives[class_name] += 1
                     # update the ".json" file
                     with open(gt_file, 'w') as f:
                         f.write(json.dumps(ground_truth_data))
                 else:
                     # false positive (multiple detection)
                     fp[idx] = 1
         else:
             # false positive
             fp[idx] = 1

2. 关于训练热身warm up的相关内容，可看文章深度学习之“训练热身”（warm up）–学习率的设置。