Delete files ec_models.py ecdetseg/__init__.py ecpose/__init__.py ecpose/engine/edgecrafter/ecpose.py examples/bakery_donuts.jpg examples/tennis_group.jpg with huggingface_hub

ec_models.py

@@ -1,112 +0,0 @@
-"""
-Hugging Face Hub wrappers around the official EdgeCrafter modules.
-
-The class definitions below mirror the reference implementation shipped by the
-authors (see `hf_models.ipynb` in https://github.com/Intellindust-AI-Lab/EdgeCrafter)
-and the composition used in `engine/edgecrafter/modeling.py` / `engine/edgecrafter/ecpose.py`.
-The `deploy()` re-parameterisation follows `tools/inference/torch_inf.py`.
-"""
-
-import torch
-import torch.nn as nn
-from huggingface_hub import PyTorchModelHubMixin
-
-from ecdetseg.engine.edgecrafter.decoder import ECTransformer
-from ecdetseg.engine.edgecrafter.ecvit import ViTAdapter as DetViTAdapter
-from ecdetseg.engine.edgecrafter.hybrid_encoder import HybridEncoder as DetHybridEncoder
-from ecdetseg.engine.edgecrafter.postprocessor import PostProcessor
-from ecpose.engine.edgecrafter.detrpose_postprocesses import DETRPosePostProcessor
-from ecpose.engine.edgecrafter.detrpose_transformer import DETRTransformer
-from ecpose.engine.edgecrafter.ecvit import ViTAdapter as PoseViTAdapter
-from ecpose.engine.edgecrafter.hybrid_encoder import HybridEncoder as PoseHybridEncoder
-
-
-def _deploy(module: nn.Module) -> nn.Module:
-    """`_ECBase.deploy()` from the reference repo: eval + re-parameterisation."""
-    module.eval()
-    for m in module.modules():
-        if hasattr(m, "convert_to_deploy"):
-            m.convert_to_deploy()
-    return module
-
-
-class _ECHubBase(nn.Module, PyTorchModelHubMixin):
-    def deploy(self):
-        _deploy(self.backbone)
-        _deploy(self.encoder)
-        _deploy(self.decoder)
-        self.postprocessor.deploy()
-        self.eval()
-        return self
-
-
-class ECDet(_ECHubBase):
-    """Object detection. Returns (labels, boxes, scores)."""
-
-    def __init__(self, config):
-        super().__init__()
-        self.config = config
-        cfg = dict(config["backbone"])
-        cfg["skip_load_backbone"] = True  # weights come from the checkpoint below
-        self.backbone = DetViTAdapter(**cfg)
-        self.encoder = DetHybridEncoder(**config["encoder"])
-        self.decoder = ECTransformer(**config["decoder"])
-        self.postprocessor = PostProcessor(**config["postprocessor"])
-
-    def forward(self, x, orig_target_sizes):
-        x = self.backbone(x)
-        x = self.encoder(x)
-        x = self.decoder(x)
-        return self.postprocessor(x, orig_target_sizes)
-
-
-class ECSeg(_ECHubBase):
-    """Instance segmentation. Returns (labels, boxes, scores, masks)."""
-
-    def __init__(self, config):
-        super().__init__()
-        self.config = config
-        cfg = dict(config["backbone"])
-        cfg["skip_load_backbone"] = True
-        self.backbone = DetViTAdapter(**cfg)
-        self.encoder = DetHybridEncoder(**config["encoder"])
-        self.decoder = ECTransformer(**config["decoder"])
-        self.postprocessor = PostProcessor(**config["postprocessor"])
-
-    def forward(self, x, orig_target_sizes):
-        x = self.backbone(x)
-        x = self.encoder(x)
-        # `modeling.ECSeg` feeds the highest-resolution encoder feature to the
-        # mask branch as `spatial_feat`.
-        x = self.decoder(x, None, x[0])
-        return self.postprocessor(x, orig_target_sizes)
-
-
-class ECPose(_ECHubBase):
-    """Multi-person 2D pose. Returns (scores, labels, keypoints)."""
-
-    def __init__(self, config):
-        super().__init__()
-        self.config = config
-        cfg = dict(config["backbone"])
-        cfg["skip_load_backbone"] = True
-        self.backbone = PoseViTAdapter(**cfg)
-        self.encoder = PoseHybridEncoder(**config["encoder"])
-        self.decoder = DETRTransformer(**config["decoder"])
-        self.postprocessor = DETRPosePostProcessor(**config["postprocessor"])
-
-    def forward(self, x, orig_target_sizes):
-        x = self.backbone(x)
-        x = self.encoder(x)
-        x = self.decoder(x, None)
-        return self.postprocessor(x, orig_target_sizes)
-
-
-TASK_CLASSES = {"detection": ECDet, "segmentation": ECSeg, "pose": ECPose}
-
-
-@torch.no_grad()
-def load_model(task: str, repo_id: str, device: str = "cpu"):
-    model = TASK_CLASSES[task].from_pretrained(repo_id)
-    model.deploy()
-    return model.to(device)

ecpose/engine/edgecrafter/ecpose.py

@@ -1,52 +0,0 @@
-
-# EdgeCrafter: Compact ViTs for Edge Dense Prediction via Task-Specialized Distillation
-# Copyright (c) 2026 The EdgeCrafter Authors. All Rights Reserved.
-# ------------------------------------------------------------------------
-# Modified from DETRPose: Real-time end-to-end transformer model for multi-person pose estimation
-# (https://github.com/SebastianJanampa/DETRPose)
-# ------------------------------------------------------------------------
-# Modified from Conditional DETR model and criterion classes.
-# Copyright (c) 2021 Microsoft. All Rights Reserved.
-# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
-# ------------------------------------------------------------------------
-# Modified from DETR (https://github.com/facebookresearch/detr)
-# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
-# ------------------------------------------------------------------------
-# Modified from Deformable DETR (https://github.com/fundamentalvision/Deformable-DETR)
-# Copyright (c) 2020 SenseTime. All Rights Reserved.
-# ------------------------------------------------------------------------
-
-from torch import nn
-
-from ..core import register
-
-__all__ = ['ECPose', ]
-
-@register()
-class ECPose(nn.Module):
-    __inject__ = ['backbone', 'encoder', 'decoder',]
-
-    def __init__(
-        self, 
-        backbone, 
-        encoder, 
-        decoder
-        ):
-        super().__init__()
-        self.backbone = backbone
-        self.encoder = encoder
-        self.decoder = decoder
-
-    def deploy(self):
-        self.eval()
-        for m in self.modules():
-            if hasattr(m, "convert_to_deploy"):
-                m.convert_to_deploy()
-        return self
-
-    def forward(self, samples, targets=None):
-        feats = self.backbone(samples)
-        feats = self.encoder(feats)
-        out = self.decoder(feats, targets, samples if self.training else None)
-        return out
-