I built an object detection model using AWS built-in algorithm: SSD with base network Resnet50 and shape of training images 300x300.
I would now like to load the model on gluon.
Would somebody know how to do it?
I tried to load the params file that was produced by the AWS training job in a gluon model that I downloaded from the model zoo (obviously not pre-trained).
The thing is, I am not sure if the zoo contains the exact model that would fit the AWS built-in algorithm.
Although the image size seems to be 512 and not 300, I still tried with ‘ssd_512_resnet50_v1_voc’ and ‘ssd_512_resnet50_v1_coco’ and ‘ssd_512_resnet50_v1_custom’ but I get the following error:
AssertionError: Parameter 'resnetv10_conv0_weight' is missing in file: trained-model-resnet-2-2002/deploy_model_algo_1-0000.params, which contains parameters: 'ssd2__plus15_loc_pred_conv_bias', 'ssd2_bn0_moving_var', 'ssd2_multi_feat_4_conv_3x3_relu_cls_pred_conv_bias', ..., 'ssd2_stage3_unit5_conv2_weight', 'ssd2_stage3_unit1_bn1_moving_var', 'ssd2_stage4_unit2_bn3_beta', 'ssd2_stage3_unit1_bn1_beta'. Please make sure source and target networks have the same prefix.
Thanks a lot for your help!