# Tracking without bells and whistles

## ICCV2019, MOTS

Posted by pshow on July 1, 2020

## Tracking without bells and whistles

### 主要贡献

• introduce the Tracktor which tackles multi-object tracking by exploiting the regression head of a detector to perform temporal realignment of object bounding boxes.
• present two simple extensions to Tracktor, a reidentification Siamese network and a motion model.
• conduct a detailed analysis on failure cases and challenging tracking scenarios, and show none of the dedicated tracking methods perform substantially better than our regression approach.
• propose our method as a new tracking paradigm which exploits the detector and allows researchers to focus on the remaining complex tracking challenges

### 简介

If a detector can solve most of the tracking problems, what are the real situations where a dedicated tracking algorithm is necessary?

### A detector is all you need

• 不需要对追踪进行特化训练
• 在测试时不需要复杂的优化过程

#### Object detector

The detector yields the final set of object detections by applying non-maximum-suppression (NMS) to the refined bounding box proposals.

#### Tracktor

$T_k={b^k_{t_1}, b^k_{t_2},…}$

$b^k_t=(x,y,w,h)$

$B_t={b^{k_1}_t, b^{k_2}_t,…}$

$D_0={d^1_0, d^2_0,…}=B_0$

Bounding box regression.

• 当分类分数$s_t^k$低于阈值$\sigma_{active}$，我们认为目标离开视野，或者是目标被非目标物体遮挡。
• 目标之间的相互遮挡，在对$B_t$进行NMS时出现IoU大于阈值$\lambda_{active}$

Bounding box initialization.

#### Tracking extensions

Motion model.

• 对于移动的摄像机we apply a straightforward camera motion compensation (CMC) by aligning frames via image registration using the Enhanced Correlation Coefficient (ECC) maximization as introduced in [16].
• 对于低帧率的视频，we apply a constant velocity assumption (CVA) for all objects as in [11, 2].

Re-identification.

### 实验

private detections和public detections版本

bounding box regressor 和 classifier 用的自己的网络。

reID和CMC作用很大。

#### 跑分测试

We evaluate the performance of our Tracktor++ on the test set of the respective benchmark, without any training or optimization on the tracking train set.

Note, we use the same Tracktor++ tracker, trained on MOT17Det object detections, for all benchmarks

### 分析

#### 对追踪场景的分析

Object visibility.

Intuitively, we expect diminished tracking performance for object-object or object-non-object occlusions, i.e., for targets with diminished visibility.

Object size.

Robustness to detections.

Identity preservation.

#### Oracle trackers

the impact of the object detector on the killing policy and bounding box regression,

identify performance upper bounds for potential extensions to our Tracktor.