setting output=target in validation does not give AP=1.0

[hengck23]

this is not an issue but rather an observation. What i did:

• use ground truth box in json as input instance box
• disable flip test
• disable oks_nms instance re-scoring
• simply set output=target in def validate(config, val_loader, val_dataset, ...) in function.py

i get the results:

Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.975 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.990 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.988 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.971 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.980 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.988 Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.999 Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.996 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.984 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.996

---

this represent results of a perfect convolution net which have predictions same as the target ground truth. Even so, there is a drop of 0.025 in AP.

This shows that there is some still room of improvement if you have better post processing methods or better target definition.