Some fixes for yolo-v3. (#529)

* Some fixes for yolo-v3.

* Use the running stats for inference in the batch-norm layer.

* Get some proper predictions for yolo.

* Avoid the quadratic insertion.

1 files changed, 7 insertions, 0 deletions

diff --git a/candle-examples/examples/yolo-v3/extract-weights.py b/candle-examples/examples/yolo-v3/extract-weights.py
new file mode 100644
index 00000000..4e523ee6
--- /dev/null
+++ b/candle-examples/examples/yolo-v3/extract-weights.py
@@ -0,0 +1,7 @@
+def remove_prefix(text, prefix):
+  return text[text.startswith(prefix) and len(prefix):]
+nps = {}
+for k, v in model.state_dict().items():
+  k = remove_prefix(k, 'module_list.')
+  nps[k] = v.detach().numpy()
+np.savez('yolo-v3.ot', **nps)