CNN_Refine not defined in DiT
1ssb opened this issue · comments
Subhransu Sekhar Bhattacharjee (Rudra) commented
Hi, I am facing an issue with running the DFM code. It says that the DiT model seems not to have the cnn_refine defined. What is the workaround? Has anyone else faced this issue?
Input:
(dfm) Singularity> python experiment_scripts/re_results.py dataset=realestate batch_size=1 num_target=1 num_context=1 model_type=dit feats_cond=true sampling_type=simple max_scenes=10000 stage=test use_guidance=true guidance_scale=2.0 temperature=0.85 sampling_steps=50 name=re10k_inference image_size=128 checkpoint_path=files/re10k_model.pt wandb=local
Output:
...
model dit
Error executing job with overrides: ['dataset=realestate', 'batch_size=1', 'num_target=1', 'num_context=1', 'model_type=dit', 'feats_cond=true', 'sampling_type=simple', 'max_scenes=10000', 'stage=test', 'use_guidance=true', 'guidance_scale=2.0', 'temperature=0.85', 'sampling_steps=50', 'name=re10k_inference', 'image_size=128', 'checkpoint_path=files/re10k_model.pt', 'wandb=local']
Traceback (most recent call last):
File "/home/usr/DFM/experiment_scripts/re_results.py", line 77, in train
model = PixelNeRFModelCond(
File "/home/usr/EK/DFM/PixelNeRF/pixelnerf_model_cond.py", line 70, in __init__
self.enc = DiT(
File "/home/usr/DFM/PixelNeRF/transformer/DiT.py", line 271, in __init__
self.initialize_weights()
File "/home/usr/DFM/PixelNeRF/transformer/DiT.py", line 319, in initialize_weights
self.cnn_refine.weight.data.fill_(0)
File "/home/usr/.conda/envs/dfm/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1614, in __getattr__
raise AttributeError("'{}' object has no attribute '{}'".format(
AttributeError: 'DiT' object has no attribute 'cnn_refine'
Zheng Chen commented
Hi! How do you solve this problem?
Subhransu Sekhar Bhattacharjee (Rudra) commented
Magically stopped getting this error. Very weird.
Update: Reconstructed this error, make sure that the data loader is working correctly. This makes little sense because the weights should not be related to the channels of the image, maybe it is related to how the transformer is parsing the image and weights in a certain order.
…On Wed, 25 Oct, 2023, 5:16 pm thucz, ***@***.***> wrote:
Hi! How do you solve this problem?
—
Reply to this email directly, view it on GitHub
<#6 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AJWHFEHF3LT2YH6BOUDCBQLYBCVERAVCNFSM6AAAAAA6LURXNGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONZYGU4DINRTHA>
.
You are receiving this because you modified the open/close state.Message
ID: ***@***.***>