Spaces:
Runtime error
Runtime error
More printing for debugging purposes
Browse files
models/GroundingDINO/groundingdino.py
CHANGED
|
@@ -409,6 +409,7 @@ class GroundingDINO(nn.Module):
|
|
| 409 |
samples.device
|
| 410 |
)
|
| 411 |
|
|
|
|
| 412 |
one_hot_token = tokenized
|
| 413 |
|
| 414 |
(
|
|
@@ -444,6 +445,7 @@ class GroundingDINO(nn.Module):
|
|
| 444 |
|
| 445 |
bert_output = self.bert(**tokenized_for_encoder) # bs, 195, 768
|
| 446 |
|
|
|
|
| 447 |
encoded_text = self.feat_map(
|
| 448 |
bert_output["last_hidden_state"]
|
| 449 |
) # bs, 195, d_model
|
|
@@ -492,6 +494,8 @@ class GroundingDINO(nn.Module):
|
|
| 492 |
else:
|
| 493 |
exemplar_tokens = None
|
| 494 |
|
|
|
|
|
|
|
| 495 |
else:
|
| 496 |
features, poss = self.backbone(samples)
|
| 497 |
(h, w) = (
|
|
@@ -597,6 +601,8 @@ class GroundingDINO(nn.Module):
|
|
| 597 |
srcs, masks, input_query_bbox, poss, input_query_label, attn_mask, text_dict
|
| 598 |
)
|
| 599 |
|
|
|
|
|
|
|
| 600 |
# deformable-detr-like anchor update
|
| 601 |
outputs_coord_list = []
|
| 602 |
for dec_lid, (layer_ref_sig, layer_bbox_embed, layer_hs) in enumerate(
|
|
|
|
| 409 |
samples.device
|
| 410 |
)
|
| 411 |
|
| 412 |
+
print("tokenized text")
|
| 413 |
one_hot_token = tokenized
|
| 414 |
|
| 415 |
(
|
|
|
|
| 445 |
|
| 446 |
bert_output = self.bert(**tokenized_for_encoder) # bs, 195, 768
|
| 447 |
|
| 448 |
+
print("got bert output")
|
| 449 |
encoded_text = self.feat_map(
|
| 450 |
bert_output["last_hidden_state"]
|
| 451 |
) # bs, 195, d_model
|
|
|
|
| 494 |
else:
|
| 495 |
exemplar_tokens = None
|
| 496 |
|
| 497 |
+
print("got visual exemplar tokens")
|
| 498 |
+
|
| 499 |
else:
|
| 500 |
features, poss = self.backbone(samples)
|
| 501 |
(h, w) = (
|
|
|
|
| 601 |
srcs, masks, input_query_bbox, poss, input_query_label, attn_mask, text_dict
|
| 602 |
)
|
| 603 |
|
| 604 |
+
print("passed info through transformer")
|
| 605 |
+
|
| 606 |
# deformable-detr-like anchor update
|
| 607 |
outputs_coord_list = []
|
| 608 |
for dec_lid, (layer_ref_sig, layer_bbox_embed, layer_hs) in enumerate(
|