I want to modify deepstream_test3 to use my tracker , i need to get frame after detection to insert it into the tracker . my problem that i can’t get the frame .
frame_meta = pyds.glist_get_nvds_frame_meta(l_frame.data) has only meta data like frame_num,source_id,batch_id,source_frame_width ,source_frame_height and num_obj_meta.
i can’t find a documentation about getting frame using pyds after inference , please help me
thank you for replay, i readed the deepstream manuel , i tested the examples , now i want to edit them and get video frame but i can’t get it based in Frame extraction using deepstream python app , Video frame data access is not included in the alpha release of DS Python bindings. i wonder if there is another method to get it from gstreamer buffer!
i want to get farme from batched , with gst_buffer = info.get_buffer() i can get befor then i can get meta data , but the frame should be in this buffer also , it can get function to get frame from buffer , it’s true there isen’t such as function?
Hi Rabeb, our upcoming release with image data access is just around the corner. Can you share your usage requirements? Would CPU accessible numpy array in RGBA format work for you?
You can use Gst.Buffer.Map() to get the input NvBufSurface from GstBuffer. However, without bindings for that structure, you can’t get to the data planes inside from Python. You can try making your own bindings for it of course.
thank you for replay , yes if i can access to CPU numpy array in RGBA format , it will be great fro me
Please check out our new release (SDK 5.0 dev preview and Python bindings v0.9). There is a sample app that shows how to do what you are asking about: https://github.com/NVIDIA-AI-IOT/deepstream_python_apps/tree/master/apps/deepstream-imagedata-multistream