You might get surprised if you run second part of the assignment( average pooling of multiple activation per image) after running the first part ( single activation). If you do not have GPU, the first part would take about 20 minutes, while the second one - 8 hours. You can reduce the processing time by using the batch learning.
In the first part the input of vl_simplenn(net,Img), where Img is 224x224x3. In this case the output of the target layer is 1x1x4096.
On the other hand, the following can be used as well: vl_simplenn(net,ImgBuf), where ImgBuf is 224x224x3x50 ( 25 cropped and 25 flipped images). The output in this case would be 1x1x4096x50.
This saved me about 3-4 hours of processing time. I got this suggestion from TA.
You might get surprised if you run second part of the assignment( average pooling of multiple activation per image) after running the first part ( single activation). If you do not have GPU, the first part would take about 20 minutes, while the second one - 8 hours. You can reduce the processing time by using the batch learning.
In the first part the input of vl_simplenn(net,Img), where Img is 224x224x3. In this case the output of the target layer is 1x1x4096.
On the other hand, the following can be used as well: vl_simplenn(net,ImgBuf), where ImgBuf is 224x224x3x50 ( 25 cropped and 25 flipped images). The output in this case would be 1x1x4096x50.
This saved me about 3-4 hours of processing time. I got this suggestion from TA.
Good luck.