Open louisquinn opened 4 years ago
I have the same question. I guess ssd_mobiledet_edgetpu_coco in TF1 model zoo is that you're looking for. Is this right?
@skn047
ssd_mobiledet_edgetpu_coco
seems to be explicitly a "Pixel4 Edge TPU model" and its mAP differs from what is written in the README (27.5 vs 25.9/25.6).
It is built from class SSDMobileDetEdgeTPUFeatureExtractor(SSDMobileDetFeatureExtractorBase)
here:
https://github.com/tensorflow/models/blob/master/research/object_detection/models/ssd_mobiledet_feature_extractor.py#L533
I'm asking about the trained model that would be built from class SSDMobileDetGPUFeatureExtractor(SSDMobileDetFeatureExtractorBase)
here:
https://github.com/tensorflow/models/blob/master/research/object_detection/models/ssd_mobiledet_feature_extractor.py#L561
Yes, same question! The sample config file for it exists here: https://github.com/tensorflow/models/blob/master/research/object_detection/samples/configs/ssdlite_mobiledet_gpu_320x320_coco_sync_4x4.config
, but there are no weights for it in the model zoo here! (even though there are pretrained weighted for every other MobileDet variant including TPU, DSP, and CPU...)
It's been more than a year since this question was posted. Are there any updates regarding the MobileDet GPU pre-trained model?
@paldana-ISI , I happen to find pre-trained weights in below website: https://coral.ai/models/object-detection/ - SSDLite MobileDet Pre-trained weights
There is the same model in tf1 detection model zoo, here
@MuttuL thanks, but I'm looking for the GPU version of the MobileDet. They only have the CPU and the EdgeTPU versions in those links you've provided.
@paldana-ISI I see that there is a directory fp32
, which i think is Float model (GPU version).
Hope it helps.
Hi all,
I can't seem to find the pre-trained SSDLite with MobileDet GPU model in both the TF1 and TF2 model zoos as referenced by the README. Are there plans to release it?
Thanks :)