OBJ2TEXT implementation with PyTorch

方法、demo及源码，参考http://www.cs.virginia.edu/~xy4cm/obj2text/

Download MSCOCO dataset including annotations

train2014/, val2014/, test2014/ and annotations/captions_train2014.json, annotations/captions_val2014.json

Collect object detection results on MS COCO dataset

• Checkout the yolo-coco-result branch of this repositoryhttps://github.com/xuwangyin/darknet/tree/yolo-coco-result, compile the code and download the weight file named yolo.weights according to https://pjreddie.com/darknet/yolo/, or just run this command wget https://pjreddie.com/media/files/yolo.weights
• List absolute paths of all the MS COCO image files into a text file find pwd -name "*.jpg" > image_files.txt, where pwd is the directory that contains MS COCO image dirs train2014, val2014 and test2014.
• Put image_files.txt in the darknet proejct directory and run ./darknet detect cfg/yolo.cfg yolo.weights data/dog.jpg > coco_detection_result to collect the result. I think it should be image_files.txt instead of data/dog.jpg.

Now we get coco_detection_result file, dectection result of coco dataset images using yolo.

Image caption

Clone https://github.com/xuwangyin/pytorch-tutorial/tree/master/tutorials/03-advanced/image_captioning.

Processing

pythob build_vocab.py --caption_path