Hello, Support team.
I’ve already run NVIDIA VSS and tested it with a sample traffic video. But in both the summary and Q&A, it’s not detecting any collisions or accidents. Could you help me figure this out?
Sincerely
Taopik H.
Sorry for the late reply. We will use the video you provided and try that on our side first.
Thank you very much.
I’ve already rented 8 * A100 (80 GB) GPUs and deployed everything using Helm.
When I deployed, I used an overrides.yaml file like this:
nim-llm:
env:
- name: NVIDIA_VISIBLE_DEVICES
value: “0,1,2,3”
resources:
limits:
nvidia.com/gpu: 0 # no limit
vss:
applicationSpecs:
vss-deployment:
containers:
vss:
env:
- name: VLM_MODEL_TO_USE
value: nvila
- name: MODEL_PATH
value: “git:Efficient-Large-Model/NVILA-15B · Hugging Face”
- name: NVIDIA_VISIBLE_DEVICES
value: “4,5”
resources:
limits:
nvidia.com/gpu: 0 # no limit
nemo-embedding:
applicationSpecs:
embedding-deployment:
containers:
embedding-container:
env:
- name: NGC_API_KEY
valueFrom:
secretKeyRef:
key: NGC_API_KEY
name: ngc-api-key-secret
- name: NVIDIA_VISIBLE_DEVICES
value: ‘6’
resources:
limits:
nvidia.com/gpu: 0 # no limit
nemo-rerank:
applicationSpecs:
ranking-deployment:
containers:
ranking-container:
env:
- name: NGC_API_KEY
valueFrom:
secretKeyRef:
key: NGC_API_KEY
name: ngc-api-key-secret
- name: NVIDIA_VISIBLE_DEVICES
value: ‘7’
resources:
limits:
nvidia.com/gpu: 0 # no limit
I used the NVILA-15B model, but when I test it, it doesn’t detect any accidents accurately.
This is a problem. Please help me with this.
Sincerely,
Taopik H.
I have tried that with the highres nvila model on my side. It can detect the accident. Could you try that?
- name: MODEL_PATH
value: "ngc:nvidia/tao/nvila-highres:nvila-lite-15b-highres-lita"
Could you please provide me with the detailed overrides.yaml file? Also, let me know which GPU(s) you rented.
And according to the documentation, I can only see the alert API for live streaming. I also need alert APIs for both video and picture summarization. Could you assist with this as well?
Sincerely,
Taopik H
override.yaml (1.4 KB)
You can create alerts for any sources.
Thanks for providing detail scripts.
But regarding alert api, what I want to know is when I summarize video or image, how can I get alert result if I use which apis? could you please detail guide for getting alert using api?
I attached alert and summarization apis. But I am not clear how can get alert result.
Please help me.
Sincerely
Taopik H.
Hello,
I have deployed highres nvila model.
Summary shows accident but Q&A doesnt show accidents. And It is not Generating Video Highlight.
Could you help this problem?
Sincerely,
Taopik H.
Hello,
Please help me with detecting accidents on Q&A.
Sincerely,
Taopik H.
I have tried that on my side.
-
click the button below
-
input the “accident” in the input field.
The highlight clips can be generated normally.
Thanks for responding.
I have deployed highres nvila model.
Summary shows accident but Q&A sometimes doesnt show accidents. It is inconsistent.
Could you help this problem?
And is it possible to deploy VILA1.5-40B model not highres in 8 * A100 GPU?
I have tried that on my side. It works normally. Could you try to delete the video and summarize that again?
There are multiple accidents but it doesn’t show.
Accuracy is missing.
Could you tell me the reason and please give me the solution.
Hello,
I’ve deployed the highres nvila model, but it’s not detecting all accidents—only a few are being recognized, and the overall accuracy is lacking. I believe the issue might be related to the model itself.
Could you please advise me on how to improve detection accuracy or how I can get more complete accident detection, on Q&A? I’d really appreciate your help.
Taopik H.
There are two methods that can be tried.
1.You need to finetune your model according to your specific scenario yourself.
2. .You can try more detailed prompts for your scenario.