README.md

Configuration

The top level of the config yaml file has

• data, see the details.
• output_dir ... Single element.
• evaluation, see the details.
• common_params, see the details
• solver, see the details
• method ... Frame-based estimation method
• params_opencv_flow ... Optical flow GT estimation parameters.
• params_openpiv ... Not used in the paper.

data

Field	Example Value	Required	Description
root		True	Root data irectory. If empty, it will be ./datasets/.
dataset	CCS	True	CSS or DAVIS
sequence	c_2022-10-20_18-27-27_41-roi-refr-100	True	The name of the sequence.
height	720	True	Height of event camera.
width	1280	True	Width of event camera.
warp	True	Optional	True to warp frames into events.

evaluation

Field	Example Value	Required	Description
metrics		No	(Not used yet)
time_list		Yes	The list (you can specify multiple) of [start_time, end_time].

common_params

Field	Example Value	Required	Description
n_frames	1	Yes	How many frames to be evaluated (or estimated), for 120 fps Basler camera. 1 means the evaluation is run based on each frame.
xmin	210	Yes	Height direction, top of the ROI. Events will be filtered accordingly.
xmax	510	Yes	Height direction, bottom of the ROI.
ymin	300	Yes	Width direction, left of the ROI.
ymax	500	Yes	Width direction, right of the ROI.

solver

Probably under method section is important, when you want to try new algorithms.

Field	Example Value	Required	Description
filter			.
filters	[BAF		List of the filters you want. Available values are [BAF, HOT]. Crop is automatically added by default, using common_params.
parameters			Parameters for the each filter.
BAF_dt	0.005		In second. Applicable when you use BAF in the filters.
BAF_ksize	3		Applicable when you use BAF in the filters.
BAF_num_support_event	2		Applicable when you use BAF in the filters.
BAF_continuous_update	True		Applicable when you use BAF in the filters.
HOT_thresh	10		Applicable when you use HOT in the filters.
method	contrast_maximization	Yes	The method of optimization. Supported values are here
warp_direction	first		Applicable when CMax-based method is used. Choose [first, middle, last].
outer_padding	0		Lorem ipsum dolor sit amet, consecteteur adipiscing.
motion_model	2d-translation		Applicable when CMax-based method is used. Supported vallues are here
parameters
cost	hybrid		Basically hybrid is fine, but you can choose from here
cost_with_weight			Applicable when cost is hybrid. Then, the following parameters will be effective.
cost_name	1.0		Effective when cost is hybrid. The cost name as the key, the weight (for the weighed sum of the total cost) as the value. Can be multple.
iwe
method	bilinear_vote		Method to create IWE, usually bilinear_vote is fine.
blur_sigma	3		Sigma for IWE. Usually 1 to 3. For BOS data I'm using 2 or 3.
optimizer
method	optuna		Optimizing method for the solver. Choose from optuna, scipt-based ones, and torch-based ones. The scipt-based and the torch-based ones are in here. They are gradient-based methods and optuna is sampling method.
sampler	uniform		Applicable when method is optuna. TPE (intelligent), uniform, and random
n_iter	800		Applicable when method is optuna. The number of sampling.
parameters			The key of this section should match with the contents of parameters
trans_x
min	-30		Min of the search area. This is displacement. Let's say you set n_frames in the common_params. Then, the -30 is the pixel within the 8.3 milliseconds (1/120).
max	30		Max of the search area.
trans_y
min	-200
max	200

method

This is the method for frame-based optical flow.

Field	Example Value	Required	Description
method	opencv_flow		all, opencv_flow, openpiv

params_opencv_flow

Field	Example Value	Required	Description
pyr_scale	0.5
levels	4
winsize	10
iterations	3
poly_n	5
poly_sigma	1.2
flags	0

params_openpiv

Field	Example Value	Required	Description
deformation_method	symmetric
windowsizes
overlap
MinMax_U_disp
MinMax_V_disp