Deepstream 5.1 python ssd mobilenet x86 #11
Conversation
There was a problem hiding this comment.
Thanks @undefined-references. it is a great feature.
I just left some comments :)
| cd ssd_mobilenet_v2 && mkdir -p 1 | ||
| cp /repo/deepstream-data/frozen_inference_graph.pb 1/model.graphdef | ||
|
|
||
| label_file_path="/repo/deepstream/5.1/ssd_mobilenet_v2/labels.txt" |
There was a problem hiding this comment.
Since Adaptive Learning API return a label_map.pbtxt, I would get label_map.pbtxt from user and parse and save it to required format.
for parsing this file you can use these modules:
https://github.com/neuralet/adaptive-object-detection/blob/main/utils/parse_label_map.py
https://github.com/neuralet/adaptive-object-detection/blob/main/protos/label_map_pb2.py
| fi | ||
|
|
||
| cd .. | ||
| python3 deepstream_ssd_parser.py $videoPath $label_file_path |
There was a problem hiding this comment.
It would be nice if you get an output_directory argument and save the output video in this directory
| fi | ||
|
|
||
| videoPath="file://$videoPath" | ||
|
|
There was a problem hiding this comment.
Can you please add another argument which takes the frozen_inference_graph.pb path from user. if the user doesn't provide a file we can download default COCO model from TensorFlow.
There was a problem hiding this comment.
Added. Thanks
| MIN_BOX_HEIGHT = 32 | ||
| TOP_K = 20 | ||
| IOU_THRESHOLD = 0.3 | ||
| OUTPUT_VIDEO_NAME = "./out.mp4" |
There was a problem hiding this comment.
Based on provided output_directory and the file name you can create an unique address for the output video path:
for example if the user provides /home/deepstream_output/ as the output directory and the input video file is softbio.mp4 the OUTPUT_VIDEO_NAME will be /home/deepstream_output/softbio_neuralet_deepstream.mp4
| MIN_BOX_WIDTH, MIN_BOX_HEIGHT) | ||
| nms_param = NmsParam(TOP_K, IOU_THRESHOLD) | ||
|
|
||
| label_names = get_label_names_from_file("ssd_mobilenet_v2/labels.txt") |
There was a problem hiding this comment.
Please don't use hard coded path. Put the input argument instead even if this is a constant address.
There was a problem hiding this comment.
modified, thanks.
| # allocated string. Use pyds.get_string() to get the string content. | ||
| id_dict = { | ||
| val: index | ||
| for index, val in enumerate(get_label_names_from_file("ssd_mobilenet_v2/labels.txt")) |
There was a problem hiding this comment.
Please don't use hard coded path. Put the input argument instead even if this is a constant address.
Improvements:
Please don't merge until I push modifications to documents.