How to output the Output6 of edge-maps in DexiNed-TF2

[zongtianhu]

I prefer the Output6 of edge-maps to the average or fused edge detection results o. What should I do in DexiNed-TF2 code? Looking forward to your reply

[xavysp]

Hi, without another training, you just need to get the output6 from res_preds output6=res_preds[5]. But if you need to train, just for the output6, it will be fast than the normal proposal because you will apply the loss function just to the output 6 can you can comment the fuse layer and other outputs.

[zongtianhu]

Thank you very much for your prompt reply.I have another question. Eight images are read from the result h5 file. According to my inference, these 8 images should represent Output1~Output6, Fuesd and Averaged in Edge-maps, as shown in Figure 1 below. . But when comparing the Edge-maps in Figure 5 in your article, you find that there are many differences between them, such as the red box in Output6. Is this difference in my understanding of the resulting h5 file or is this a normal difference in DexiNed every time it trains? Looking forward to your reply again.

[xavysp]

Thank you very much for your prompt reply.I have another question. Eight images are read from the result h5 file. According to my inference, these 8 images should represent Output1~Output6, Fuesd and Averaged in Edge-maps, as shown in Figure 1 below. . But when comparing the Edge-maps in Figure 5 in your article, you find that there are many differences between them, such as the red box in Output6. Is this difference in my understanding of the resulting h5 file or is this a normal difference in DexiNed every time it trains? Looking forward to your reply again.

Hi, its a great question, did you make a quantitative evaluation and compared those results? Usually when you use CNN model from different training you will find a slightly difference, but the quantitative results should be almost the same. You know, you can find two checkpoints train1 and train 2, the train one is for the WACV, the second one is with the last annotation of BIPED.

Hope I answered your question, otherwise I will be happy to reply.

Cheers,

Xavier