insectdct-csv
All format examples · Machine-readable schema · Convert outputs to COCO
One source-example row from sequence-CL.csv. This multi-table profile has other layouts described in its schema notes. CL has pixel boxes and classifier scores in percent; HI contains hierarchical details.
Example
Source-example row retained in the schema notes; not a newly generated prediction.
trap,trapId,date,time,taxaConf,taxaLabel,taxaId,taxaLevel,frameId,x1,y1,x2,y2,fileName
sequence,0,20250221,115731,0.03140208778682231,Apoidea striped,12,3,0,1307,601,1355,663,sequence/pi2_2025_02_21_11_57_31.jpg
Convert to a standard format
This format is documented, but a conversion adapter is not yet registered.
General conversion example and required context
How to interpret it
Scope: Default CSVformat=tracking and hierarchical ConvNextBase V6. Each row is a detected/classified instance. Filename stems derive from the input sequence.
Coordinates: CL x1,y1,x2,y2 are pixel corners in the processed frame; frame preprocessing and motion alignment must be retained with the results.
Confidence: CL taxaConf is classifier probability multiplied by 100. HI Conf1/2/3 contain classifier probabilities, not YOLO detector scores.
Classes: CL taxaId adds one to the selected classifier index. HI LabelId fields retain classifier indices. These are not the detector class vocabulary.
Representation: Each structure variant describes one CSV table after reading the header as field names and decoding numeric cells. The producer writes CSV, not JSON objects. notes.tables gives each filename, exact column order and an observed example row; variable filename stems are described in scope. Empty cells remain empty strings, not absent fields.