Skip to content

Latest commit

 

History

History
81 lines (74 loc) · 2.66 KB

Dataset_Instructions.md

File metadata and controls

81 lines (74 loc) · 2.66 KB
Raw

Instructions

Our SEE dataset provides both intensity frames and accumulated event frames. The event accumulation period is 33 ms, corresponding to the event-based camera's ASP frame rate, i.e., 30 FPS.

The file structure is shown as follows:

SEE/
|-- event
|   |-- angry
|   |   |-- 1_001_man_24_Master_normal_angry_45_take000
|   |   |   |-- 00000.jpg
|   |   |   |-- 00001.jpg
|   |   |   |-- ...
|   |   |   |-- ...
|   |   |   |-- ...
|   |   |-- 2_001_man_24_Master_normal_angry_50_take001
|   |   |-- ...
|   |   |-- ...
|   |   `-- ...
|   |-- disgust
|   |-- fear
|   |-- happiness
|   |-- neutraL
|   |-- sadness
|   `-- surprise
`--  frame
    |-- angry
    |-- disgust
    |-- fear
    |-- happiness
    |-- neutraL
    |-- sadness
    `-- surprise

Sequence Name Format

For each sequence, there are two folders: one is under the frame folder, containing the intensity frames; the other is under the event folder, providing the corresponding accumulated event frames. The two folders have the same name, with the following format:

[sequence#]_[ID]_[sex]_[age]_[degree]_[lighting-condition]_[emotion]_[length]_[take#]

[sequence#]: the sequence number [ID]: volunteer ID [sex]: volunteer's sex, man or woman [age]: volunteer's age [degree]: volunteer's highest educational degree [lighting-condition]: the lighting condition of the sequence [emotion]: the ground truth label of the sequence [length]: the length of the sequence in the number of frames [take#]: the sequence is the [take#]th sequence of the volunteer with [ID]

For example:

1_001_man_24_Master_normal_angry_45_take000

[sequence#=1]: the first sequence [ID=001]: volunteer ID is 001 [sex=man]: a male volunteer [age=24]: volunteer's age is 24 years old [degree=Master]: volunteer's highest educational degree is a master degree [lighting-condition=normal]: the sequence is recorded under normal lighting condition [emotion=angry]: the ground truth label of the sequence is angry [length=45]: total number of frames in the sequence is 45 [take#]: It is the first sequence of the volunteer 001.

Dataset Training/testing split

The training/testing split is provided in the emotion.json One example is shown as follows:

"1011_035_man_20_undergraduate_normal_angry_63_take000": {
			"subset": "training",
			"annotations": {
				"label": "angry",
				"segment": [0, 63]
			}
		},

[subset: training]: indicates the seqeuce belongs to training set [label: angry]: indicates ground truth label of the sequence is angry [segment: [0,63] ]: indicates the start and end frame index of the sequence