How to add (weight, probability and magnitude) to the forward calculation to calculate the gradient？ about dada HOT 3 OPEN

The gradient is auto-calculated when backward, since the use of images.detach() + adds.
Two paths exist.
In the first path, the gradient will be backward to com_image, then to images, then to adds, then to probability[i] and magnitude[i].
In the second path, the gradient will be backward to w, weights, op_weights.

from dada.

hello, I have some question in search_relax.

parser.add_argument('--use_parallel', type=bool, default=False, help="use data parallel default False")
The default valse is False, and in
def forward(self, origin_images, probabilities_b, magnitudes, weights_b):
the args are probabilities_b, magnitudes, weights_b, how to calculate the gradient of model.probabilities and model.ops_weights

from dada.

@Trent-tangtao did you resolve this problem？ how to calculate the gradient of model.probabilities and model.ops_weights

from dada.