Research Article
Object Detection Based on Fast/Faster RCNN Employing Fully Convolutional Architectures
Table 2
The detailed feature size of each layer in the Inception_v3 architecture.
| Layers | Input size | Pad | Kernel size | Output size |
| Conv1-s2 | ( = 5~12) | 0 | | ( = 18~21) | Conv2 | ( = 18~21) | 0 | | ( = 16~19) | Conv3 | ( = 16~19) | 1 | | ( = 16~19) | MaxPool-s2 | ( = 16~19) | 0 | | or | Conv4_1 | or | 0 | | or | Conv4_2 | or | 0 | | or | MaxPool-s2 | or | 0 | | | Inc_a1~Inc_a3 | | 1 | | | Inc_a-s2 | | | | | Conv | | 0 | | | Pool | | 0 | | | Inc_b1~Inc_b4 | | 1 | | | Inc_b-s2 | | | | | Conv | | 0 | | | Pool | | 0 | | | Inc_c1~Inc_c2 | | 1 | | |
| GAP |
| Fc_1000 |
| Softmax |
|
|