Skip to content

align train_step with new object dict structure and loss input format #16

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
bernardev254 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

align train_step with new object dict structure and loss input format #16

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 11 additions & 4 deletions hackathon/objectdetection.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -133,12 +133,19 @@ def loss_fn(params, static, images, objects, key):

@eqx.filter_jit
def train_step(params, static, optimizer, opt_state, batch, key):
images = batch["global_crops"][:, 0, :, :, :]
objects = batch["objects"][:, 0, :, :, :]
classes = batch["classes"]
images = batch["global_crops"][:, 0] # shape: (B, H, W, C)
images = nhwc_to_nchw(images) # → (B, C, H, W)

object_data = batch["objects"] # dict of lists/tensors

targets = {
"bbox_target": object_data["bboxes"], # shape: (B, H, W, 4)
"class_target": object_data["labels"], # shape: (B, H, W, C)
"object_mask": object_data["object_mask"], # shape: (B, H, W, 1)
}

def loss_for_step(p):
return loss_fn(p, static, images, objects, classes, key)
return loss_fn(p, static, images, targets, key)

loss, grads = jax.value_and_grad(loss_for_step)(params)

Expand Down