I trained the Tensorflow Object Detection Centernet model with Hourglass backbone. The checkpoint size is 1.53 GB. Is the checkpoint size normal and not too large? The Faster Rcnn model with Resnet50 backbone, which is trained on the same dataset, has a checkpoint with 227 MB.
Related
I want to train YOLOv3 Darknet on a custom dataset using MS COCO pre-trained weights. Specifically, i want to fine-tune the MS COCO pre-trained weights of the YOLOv3 on a custom dataset.
I followed the instructions of data preparation and detector configuration in the AlexeyAB/darknet Github repository, but in pre-trained weights part, only the use of ImageNet pre-trained convolutional weights was explained. I googled about fine-tuning MS COCO pre-trained weights of the YOLOv3 on a custom dataset, but couldn't found any explanation.
I trained ssd mobilenet v1 on custom dataset now I want to run in jetson, I converted it to frozen graph pb file using tensorflow object detection api, i want to run this model on jetson nano, but I eats 2.5 GB of RAM, while model size is only 22.3MB. I tried with Tensorrt FP16 conversion and still same memory consumption.
I need that model in the size of 5 to 6 MB or at least it must consume less memory on inference.
How can I load coco keypoints dataset (image and keypoint annotations) in tensorflow using tf.data.Dataset?
There's a coco Dataset in tensorflow but it's only for object detection.
Regards.
I have a frozen inference graph(frozen_inference_graph.pb) and a checkpoint (model.ckpt.data-00000-of-00001, model.ckpt.index), how to deploy these to Tensorflow serving? serving need SavedModel format, how to convert to it?
I study Tensorflow and found Deeplab v3+ provide PASCAL VOC 2012 model, I run train, eval, visualization on my local PC, but I don't know how to deploy it on serving.
Have you tried export_inference_graph.py?
Prepares an object detection tensorflow graph for inference using model
configuration and a trained checkpoint. Outputs inference
graph, associated checkpoint files, a frozen inference graph and a
SavedModel
I have found TPU compatible pre-trained checkpoints for object detection but not for image classification. I'm specifically interested in Mobilenet_v2. I would switch to another network if such checkpoints were available and those for Mobilenet_v2 were not.
Yes, there is a TPU-compatible Resnet checkpoint in these GCS files:
gs://cloud-tpu-artifacts/resnet/resnet-nhwc-2018-02-07/checkpoint
gs://cloud-tpu-artifacts/resnet/resnet-nhwc-2018-02-07/model.ckpt-112603.data-00000-of-00001
gs://cloud-tpu-artifacts/resnet/resnet-nhwc-2018-02-07/model.ckpt-112603.index
gs://cloud-tpu-artifacts/resnet/resnet-nhwc-2018-02-07/model.ckpt-112603.meta
You can use this checkpoint if it meets your requirements. If not, you can follow this tutorial to train it yourself.