我们要感谢 Kaggle 和 Sartorius 组织了这样一场精彩的比赛。

概述

我们的解决方案如下。在许多方面，它与第二名的解决方案相似。

在比赛一开始，我们决定构建一个基于边界框的实例分割解决方案，并将更多精力集中在边界框检测部分。我们认为掩码预测的性能主要受限于标注质量，因此我们没有对此投入过多精力。
在比赛期间，我们使用 COCO mAP 作为验证指标，我们相信高 mAP 和适当的阈值处理会带来高 LB 分数。以下是我们在比赛结束时取得的验证分数：

评估 bbox...
加载并准备结果...
完成 (t=0.09s)
创建索引...
索引创建完成！

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=2000 ] = 0.396
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=2000 ] = 0.764
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=2000 ] = 0.364
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=2000 ] = 0.354
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=2000 ] = 0.305
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=2000 ] = 0.572
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.386
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=300 ] = 0.491
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=2000 ] = 0.579
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=2000 ] = 0.550
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=2000 ] = 0.552
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=2000 ] = 0.757
    

评估 segm...
加载并准备结果...
完成 (t=0.57s)
创建索引...
索引创建完成！

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=2000 ] = 0.362
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=2000 ] = 0.767
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=2000 ] = 0.294
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=2000 ] = 0.308
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=2000 ] = 0.419
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=2000 ] = 0.461
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.349
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=300 ] = 0.441
 Average Recall     (AR) @[ IoU