Replies to post #10290 on SCIENCE SPIN-OFFS

[dockzef]

Codes. We provide a Python freeware software package, called DeepTrack, which can be readily personalized and optimized for the needs of speci?c users and applications. The key functions are contained in the library “deeptrack.py”, while ?ve applications are provided in the Example Codes 1-4, each of which consists of a Jupyter notebook and some additional ?les for the pretrained neural networks (in .h5 format) and some sample videos (in .mp4 format). Example Code 1a generates some particle images like those employed in Figs. 1b-c and tracks their position using the pre-trained network “DeepTrack - Example 1a - Pretrained network.h5”. Example Code 1b demonstrates the whole training process for the neural network employed in Figs. 1b-c: from de?nition of the parameters for the simulation of the particle images, to the de?nition of the neural network, to its training, and ?nally to its testing. Example Code 2 generates some particle images like those employed to train the neural network used in Fig. 2. Furthermore, using the pre-trained neural network “DeepTrack - Example 2 - Pretrained network.h5”, it tracks the positions of the optically trapped particle in ideal (see Figs. 2a-d and Video 1) and poor (see Figs. 2eh and Video 2) illumination conditions using the corresponding original videos (“DeepTrack - Example 2 - Optically Trapped Particle Good.mp4” and “DeepTrack Example 2 - Optically Trapped Particle Bad.mp4”, respectively). Example Code 3 generates some particle images like

those employed to train the neural network used in Figs. 3 and 4a-b. Furthermore, using the pre-trained neural network “DeepTrack - Example 3 - Pretrained network.h5”, it tracks the positions of multiple particles in ideal (see Figs. 3 and Video 3) and poor (see Fig. 4a and Video 4) illumination conditions as well as bacteria (see Fig. 4b and Video 5) using the corresponding original videos (“DeepTrack - Example 3 - Brownian Particles Good.mp4”, “DeepTrack - Example 3 - Brownian Particles Bad.mp4”, and “DeepTrack - Example 3 - Bacteria.mp4”, respectively). Example Code 4 generates some particle images like those employed to train the neural network used in Fig. 4c. Furthermore, using the pre-trained neural network “DeepTrack - Example 4 - Pretrained network.h5”, it tracks the positions of multiple particles in poor illumination conditions and in the presence of bacteria (see Fig. 4c and Video 6) using the corresponding original video (“DeepTrack - Example 4 - Brownian particles + bacteria.mp4”). We thank Falko Schmidt for help with the ?gures, Giorgio Volpe for critical reading of the manuscript, and Santosh Pandit and Er¸ca?g Pinc¸e for help and guidance with the bacterial cultures. This work was supported by the ERC Starting Grant ComplexSwimmers (Grant No. 677511).