docs: add README + license for fcn_resnet50 (#1130)

README.md

@@ -0,0 +1,88 @@
+        ---
+        license: bsd-3-clause
+        tags:
+          - vision
+          - pytorch
+          - torchvision
+          - ferrotorch
+        ---
+
+        # `ferrotorch/fcn_resnet50`
+
+        FCN with ResNet-50 backbone, pretrained on a COCO subset with Pascal VOC labels (21 classes). Re-keyed from torchvision 0.21 `fcn_resnet50` (`FCN_ResNet50_Weights.COCO_WITH_VOC_LABELS_V1`).
+
+        ## Provenance
+
+        * Upstream factory: `torchvision.models.fcn_resnet50`
+          with `weights="COCO_WITH_VOC_LABELS_V1"` (torchvision 0.21).
+        * Conversion script: [`ferrotorch/scripts/pin_pretrained_weights.py`](https://github.com/dollspace/ferrotorch/blob/main/scripts/pin_pretrained_weights.py).
+        * Ferrotorch issue: <https://github.com/dollspace/ferrotorch/issues/1130>.
+        * Number of trainable parameters in upstream torchvision model: **35,322,218**.
+        * SHA-256 of `model.safetensors` (this file is pinned in
+          `ferrotorch-hub/src/registry.rs`): `8419d91ad57f4156e3a6add39abd43caf0a3761083743fe0a5dddf470ffdabf7`.
+
+        ## How to load
+
+        ```rust
+        use ferrotorch_vision::models::registry::get_model;
+        let model = get_model("fcn_resnet50", /* pretrained = */ true, /* num_classes = */ 21).unwrap();
+        ```
+
+        The loader downloads this file, verifies SHA-256, then calls
+        `Module::load_state_dict(state_dict, strict=false)`. `strict=false`
+        is required because ferrotorch's `Module::named_parameters()` does
+        not yet expose `BatchNorm2d` running statistics
+        (`running_mean` / `running_var`), so those keys in this safetensors
+        file are intentionally ignored at load time until ferrotorch
+        issue #995 closes. They are still included here so re-uploading
+        is unnecessary once that work lands.
+
+        ## Conversion notes
+
+        The following upstream torchvision keys are intentionally dropped because the ferrotorch architecture does not have a corresponding parameter slot:
+
+  * `aux_classifier.0.weight`
+  * `aux_classifier.1.bias`
+  * `aux_classifier.1.num_batches_tracked`
+  * `aux_classifier.1.running_mean`
+  * `aux_classifier.1.running_var`
+  * `aux_classifier.1.weight`
+  * `aux_classifier.4.bias`
+  * `aux_classifier.4.weight`
+
+For FPN bias drops this is a known mismatch between torchvision's `nn.Conv2d(..., bias=True)` FPN convolutions and ferrotorch's `bias=False` FPN convolutions. For `aux_classifier.*` drops the ferrotorch DeepLabV3 / FCN implementations do not expose an aux head.
+
+        ## Upstream license (verbatim, torchvision 0.21 `LICENSE`)
+
+        ```
+        BSD 3-Clause License
+
+        Copyright (c) Soumith Chintala 2016,
+        All rights reserved.
+
+        Redistribution and use in source and binary forms, with or without
+        modification, are permitted provided that the following conditions are met:
+
+        * Redistributions of source code must retain the above copyright notice, this
+          list of conditions and the following disclaimer.
+
+        * Redistributions in binary form must reproduce the above copyright notice,
+          this list of conditions and the following disclaimer in the documentation
+          and/or other materials provided with the distribution.
+
+        * Neither the name of the copyright holder nor the names of its
+          contributors may be used to endorse or promote products derived from
+          this software without specific prior written permission.
+
+        THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+        AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+        IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+        DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+        FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+        DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+        SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+        CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+        OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+        OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+        ```