Freeze block in Model config

Hi all
In model_config there is option to freeze block. i am using resnet 34 as a backbone feature extractor with freeze_block:0 and i know by using freeze block weight will not update but i am confused what is this block 0 contains . In website they mention the block ID’s valid for freezing is any subset of [0, 1, 2, 3] but what does this mean.
Please help me in this.

See https://docs.nvidia.com/metropolis/TLT/tlt-getting-started-guide/index.html#network_config_fasterrcnn

The list of block IDs to be frozen in the model during training. You can choose to freeze some of the CNN blocks in the model to make the training more stable and/or easier to converge. The definition of a block is heuristic for a specific architecture. For example, by stride or by logical blocks in the model, etc. However, the block ID numbers identify the blocks in the model in a sequential order so you don’t have to know the exact locations of the blocks when you do training. A general principle to keep in mind is: the smaller the block ID, the closer it is to the model input; the larger the block ID, the closer it is to the model output.

Thanks @Morganh. Now I am clear on that.
And in model config one more option is there called objective set :This defines what objectives is this network being trained for. For object detection networks, set it to learn cov and bbox.
so that our objective is to learn cov(coverge) and bbox but they are suggesting to use:
cov {
}
bbox { scale: 35.0 offset: 0.5
}
so, why cov is empty, why we are not giving any value for cov?
Actually i trained my model on custom dataset with the above objective set and my model is giving very less confidence while inferencing.
From my understanding,because i am not setting cov in model config that’s why confidence is low while inferencing.
Please correct me if i am wrong and what value should i give to cov.

The objective_set contains valid parameters for bbox only.
There is post processing for bbox labels, but no for cov.

For low confidence you got, please note that there are many factors result in low confidence. For example,

  • Are test image similar to training images?
  • is dataset large enough?