elegantlin / cvwc-2019

ICCV 2019 Workshop & Challenge on Computer Vision for Wildlife Conservation (CVWC)

CVWC-2019

ICCV 2019 Workshop & Challenge on Computer Vision for Wildlife Conservation (CVWC) Detection and ReID Tracks

Data Pre-processing

Before we designed our model, we looked into the dataset carefully and the details are as follows:

The distribution of bounding box according to size, count and square.

Results

We joined 3 tracks of 4 in this workshop which are detection, plain ReID and wild ReID.

The final results are:

Track	mAP(single_cam in ReID)	top-1(single_cam)	top-5(single_cam)	mAP(cross_cam)	top-1(single_cam)	top-5(cross_cam)	GFLOPs	Param
Detection	0.60	-	-	-	-	-	245.3	45.0M
Plain ReID	0.84	0.95	0.99	0.46	0.87	0.93	-	-
Wild ReID	0.82	0.93	0.96	0.46	0.84	0.90	-	-

Please refer to the respective folder for more details.

Members

• Zonglin DI
• BingChen ZHAO