Loads database and creates reverse indexes and shortcuts. :param version: Version to load (e.g. "v1.0-train", "v1.0-val", "v1.0-test"

Remove the non-conditioning memory around the input frame. When users provide correction clicks, the surrounding frames' non-conditio

Identical to the corresponding method in the parent (SAM2VideoPredictor), but cloning the outputs of prompt_encoder and mask_decoder

Map model-side object index to client-side object id.

Add new mask to a frame.

Add new mask to a frame.

Deprecated method. Please use `add_new_points_or_box` instead.

Deprecated method. Please use `add_new_points_or_box` instead.

Generate argparser from config file automatically (experimental)

Generate argparser from config file automatically (experimental)

Compute the DICE loss, similar to generalized IOU for masks Args: inputs: A float tensor of arbitrary shape. The pred

Args: inputs: A float tensor of arbitrary shape. The predictions for each example. targets: A float tensor with t

Loss used in RetinaNet for dense detection: https://arxiv.org/abs/1708.02002. Args: inputs: A float tensor of arbitrary shape.

Compute the DICE loss, similar to generalized IOU for masks Args: inputs: A float tensor of arbitrary shape. The pred

Compute the DICE loss, similar to generalized IOU for masks Args: inputs: A float tensor of arbitrary shape. The pred

preds: [bs, num_class + 1, h, w] labels: [bs, h, w]

Args: text_features: [hidden_size] 文本特征 vision_features: [vision_dim] 视觉特征（可选） Returns: enhanced_

Args: text_features: [hidden_size] 文本特征 global_latents: [num_latents, hidden_size] 全局latent Returns:

Args: x (torch.Tensor): image features shape (b, T, n1, D) latents (torch.Tensor): latent features

保持原有接口不变

Args: x (torch.Tensor): image features shape (b, T, n1, D) latent (torch.Tensor): latent features

Args: latents (torch.Tensor): query tokens shape [(b, T, n, D)] x: multi-scale image features