I use the object_detection_tutorial.ipynb default model ssd_mobilenet_v1_coco_2017_11_17 under the object_detection API. Make changes to the test picture dog.
Figure 1 is the original image, which can identify 2 dogs. I took a picture of the dog on the left side of Figure 1 and tested it as a dog. Then, Figure 2 is placed in a different position, even though Figures 3, 4, and 5 can be recognized as a dog. But why is ACC different? Figure 6 shows that I copied Figure 2 five times in a picture. Only three dogs can be identified at this time. Why are there two that are not recognized? ACC will be so low!