- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
I'm trying the Ifxdaq labeling tool with a BGT60TR13C demo board and an Intel RealSense camera.
The recording session works fine but when I'm trying the labeling session with the pre-defined labeling pipeline I've this error:
ConnectionError: Connection to Artifactory not possible. URL: https://artifactory.intra.infineon.com/artifactory/gen-pmm-sensys-local/ifxdaq/models
Can you help me to fix this error?
Best Regards,
François
Solved! Go to Solution.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi François,
running the labeling algorithms outside the Infineon company network requires currently a few manual steps - this is one of our main topics for the upcoming releases to automate this process & to enhance the user experience here.
Documentation: Labeling > Stages > Detection Stage
For the moment, let me guide you with some steps:
1. Download the model, e.g. YOLOv5s from https://github.com/ultralytics/yolov5/releases/download/v6.0/yolov5s.pt. You'll find a list of all publicly available models in the documentation under API > ifxdaq.ai.model_zoo.
Documentation: API > ifxdaq.ai.model_zoo > Ultralytics YOLOv5
2. Please place the weights inside the ifxdaq cache directory under the relative path described under API > ifxdaq.ai.model_zoo, e.g. $IFXDAQ_CACHE/ultralytics/yolov5/torchhub/yolov5s.pt
Documentation: Labeling > Stages > Detection Stage
3. Specify the selected labeling algorithm in the pipeline definition, e.g. human_tracking.yaml. (I re-used the pre-defined human_tracking pipeline with a publicly available model - see the difference in line 2.) - you will find more information in the documentation under Labeling > Usage:
- yolo_detection: model: "yolov5s" threshold: 0.75 person_only: True - tracking: [] - crop_lag: [] - merge_tracks: threshold_metric: 0.3 threshold_time: 5 - filter_short: threshold_time: 5 - world_coordinates: [] - filter_hampel: time_window: 5 n_sigma: 3 - filter_savgol: window_length: 51 polyorder: 3 - visualization: anonymization: PIXEL anonymization_type: AUTO - compose: output_name: "Labels" cleanup: True
4. Run the labeling & point to the configuration file:
ifxdaq label -c human_tracking CAMERA_RECORD...
Please let me know if this solved your issue - As mentioned above, we're working on automating this process globally 😉
Max
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi @f_raymackers ,
You are trying to access the Infineon Artifactory is only accessible from within the Infineon company network. Please try using the .whl file for installation. Do let us know if that works for you !
Best regards,
Deepa
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Thanks for your reply.
The installation was done with the following command:
pip install "ifxdaq-<version+platform>.whl[ai]"
The recording tool works like a charm. But a the labeling step, I have connecting errors when I execute the following command (command from the part "Basic workflow" of the tutorial):
ifxdaq label -p human_tracking /recording_2022_05_25_07_53_56/CamIntelRealSense_00
See the error on the attachement screen.
Thanks
Best Regards,
François
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi François,
running the labeling algorithms outside the Infineon company network requires currently a few manual steps - this is one of our main topics for the upcoming releases to automate this process & to enhance the user experience here.
Documentation: Labeling > Stages > Detection Stage
For the moment, let me guide you with some steps:
1. Download the model, e.g. YOLOv5s from https://github.com/ultralytics/yolov5/releases/download/v6.0/yolov5s.pt. You'll find a list of all publicly available models in the documentation under API > ifxdaq.ai.model_zoo.
Documentation: API > ifxdaq.ai.model_zoo > Ultralytics YOLOv5
2. Please place the weights inside the ifxdaq cache directory under the relative path described under API > ifxdaq.ai.model_zoo, e.g. $IFXDAQ_CACHE/ultralytics/yolov5/torchhub/yolov5s.pt
Documentation: Labeling > Stages > Detection Stage
3. Specify the selected labeling algorithm in the pipeline definition, e.g. human_tracking.yaml. (I re-used the pre-defined human_tracking pipeline with a publicly available model - see the difference in line 2.) - you will find more information in the documentation under Labeling > Usage:
- yolo_detection: model: "yolov5s" threshold: 0.75 person_only: True - tracking: [] - crop_lag: [] - merge_tracks: threshold_metric: 0.3 threshold_time: 5 - filter_short: threshold_time: 5 - world_coordinates: [] - filter_hampel: time_window: 5 n_sigma: 3 - filter_savgol: window_length: 51 polyorder: 3 - visualization: anonymization: PIXEL anonymization_type: AUTO - compose: output_name: "Labels" cleanup: True
4. Run the labeling & point to the configuration file:
ifxdaq label -c human_tracking CAMERA_RECORD...
Please let me know if this solved your issue - As mentioned above, we're working on automating this process globally 😉
Max
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello Max,
Thanks for your reply and your help, I think we are moving in the right direction but I have again a connection error.
I did all manipulations you described and when I launch a labeling session, I've first a detection loading then I have the same error (visible in my screen in attachment).
After the launch of the labeling, I noticed a "Detection_00" folder has been created in my directory. Is it good?
I placed in attachement the log file of the labelling.
Thanks and best regards,
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi François,
yes, we move definitely into the right direction!
There's one thing I forgot in my last post. After the detection, we apply a tracker which is based on a ReID network that needs some model weights as well. Please download the weights (https://drive.google.com/file/d/1_qwTWdzT9dWNudpusgKavj_4elGgbkUN/view?usp=sharing - you'll find the link also in the documentation) & place them inside the cache directory under :
$IFXDAQ_CACHE/DeepSORT/ReID/ckpt.t7
Documentation: Labeling > Stages > Tracking Stage
Documentation: API > ifxdaq.ai.model_zoo > ReID Models
Hope now I forgot no more steps 😉
Max
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi Max,
Perfect, the labelling works like a charm! Thank your for your detailed help 😊
Best Regards,
François