ianpan
/

bone-age

@@ -19,7 +19,7 @@ from transformers import AutoModel
 model = AutoModel.from_pretrained("ianpan/bone-age", trust_remote_code=True)
 ```
 The model is a 3-fold ensemble utilizing the `convnextv2_tiny` backbone.
-The individual models can be accessed through `model.net0`, `model.net1`, `model.net2`.
 Originally, it was trained with both a regression and classification head.
 However, this model only loads the classification head, as stand-alone performance was slightly better. The classification head also generates better GradCAMs.
 The softmax function is applied to the output logits and multiplied by the corresponding class indices, then summed.
@@ -29,8 +29,6 @@ In addition to standard data augmentation, additional augmentations were also ap
 - Using a cropped radiograph (from the model <https://huggingface.co/ianpan/bone-age-crop>) with probability 0.5
 - Histogram matching with a reference image (available in this repo under Files, `ref_img.png`) with probability 0.5
-The model was trained over 20,000 iterations using a batch size of 64 across 2 NVIDIA RTX 3090 GPUs.
 Note that both of the above augmentations could be applied simultaneously and in conjunction with standard data augamentations. Thus, the model accommodates a large range of variability in the appearance of a hand radiograph.
 On the original challenge test set comprising 200 multi-annotated pediatric hand radiographs, this model achieves a **mean absolute error of 4.16 months** (when applying both cropping and histogram matching to the input radiograph), which surpasses the [top solutions](https://pubs.rsna.org/doi/10.1148/radiol.2018180736) from the original challenge.
@@ -86,6 +84,8 @@ coords = coords[0].cpu().numpy()
 x, y, w, h = coords
 # coords already rescaled with img_shape
 cropped_img = img[y: y + h, x: x + w]
 ref = cv2.imread("ref_img.png", 0) # download ref_img.png from this repo
 cropped_img = match_histograms(cropped_img, ref)

 model = AutoModel.from_pretrained("ianpan/bone-age", trust_remote_code=True)
 ```
 The model is a 3-fold ensemble utilizing the `convnextv2_tiny` backbone.
+The individual single-fold models can be accessed through `model.net0`, `model.net1`, `model.net2`. Each of these models was trained over 20,000 iterations using a batch size of 64 across 2 NVIDIA RTX 3090 GPUs.
 Originally, it was trained with both a regression and classification head.
 However, this model only loads the classification head, as stand-alone performance was slightly better. The classification head also generates better GradCAMs.
 The softmax function is applied to the output logits and multiplied by the corresponding class indices, then summed.
 - Using a cropped radiograph (from the model <https://huggingface.co/ianpan/bone-age-crop>) with probability 0.5
 - Histogram matching with a reference image (available in this repo under Files, `ref_img.png`) with probability 0.5
 Note that both of the above augmentations could be applied simultaneously and in conjunction with standard data augamentations. Thus, the model accommodates a large range of variability in the appearance of a hand radiograph.
 On the original challenge test set comprising 200 multi-annotated pediatric hand radiographs, this model achieves a **mean absolute error of 4.16 months** (when applying both cropping and histogram matching to the input radiograph), which surpasses the [top solutions](https://pubs.rsna.org/doi/10.1148/radiol.2018180736) from the original challenge.
 x, y, w, h = coords
 # coords already rescaled with img_shape
 cropped_img = img[y: y + h, x: x + w]
+# histogram matching
 ref = cv2.imread("ref_img.png", 0) # download ref_img.png from this repo
 cropped_img = match_histograms(cropped_img, ref)