Pytoch deep learning and target detection practice notes

Original video link ：PyTorch Deep learning target detection introduction practical series 【 Mound x Boolean number 】

download VOC Data sets

Download address ：http://host.robots.ox.ac.uk/pascal/VOC/

What is in the dataset

• Annotations Label folder ( contain xml file , Various information of pictures )
• ImageSets Picture collection ( Main concern Main The contents of the folder )
• JPEGImages The data set contains pictures
• SegmentationClass Semantic segmentation picture
• SegmentationObject Instance split picture

What we need to pay attention to ImageSets

open ImageSets-Main-aeroplane_train.txt

You can see 32 and 33 by 1, stay JPEGImages in ,32 Picture of No

open Annotations

open Annotations-000032.xml

<object>
		<name>aeroplane</name>< Picture category name >
		<pose>Frontal</pose>< The angle of the object , It can be seen that it was taken in front >
		<truncated>0</truncated>< Whether it is truncated ,0 Yes no >
		<difficult>0</difficult>< Is it difficult to identify ,0 It's not difficult >
		<bndbox>
			<xmin>104</xmin>< The minimum value of abscissa of the object on the picture >
			<ymin>78</ymin>< The minimum value of the vertical coordinate of the object on the picture >
			<xmax>375</xmax>< The maximum abscissa of the object on the picture >
			<ymax>183</ymax>< The maximum value of the ordinate of the object on the picture >
				</bndbox>
</object>

download COCO Data sets

Official website ：https://cocodataset.org/#home

Reading Annotations

Scan from the first line , yes 46 Lattice pixel , White pixels do not appear until the third line , Line by line scanning
“counts”:[147,3,1…]

To be continued